Are these the same bytes?
A cryptographic digest can support an exact-byte comparison to a specific receipt or trusted reference. Hashing begins when bytes are observed; it cannot establish their earlier history or the truth of the depicted event.
What this library can tell you. File identity, source similarity, embedded marks, signed claims, capture attestations and forensic scores answer different questions. Preserve those distinctions instead of converting them into a single real or fake verdict.
Images, audio, video and their metadata. Covers major method families, public standards, representative research implementations, and documented commercial services. A broad research snapshot, not a claim to enumerate every paper, patent, private algorithm or current product.
Start with the claim
The same file can have valid provenance, match an older source, and still be presented with a false story. Read each result as evidence for a specific claim. The guide below is a synthesis; the catalog provides method-specific primary sources.
Is this a copy or derivative of something already indexed?
Perceptual hashes, local features and embeddings can retrieve candidate relationships despite some changes. Matching is threshold-dependent and bounded by the reference collection. Similarity alone does not establish ownership, a unique origin or semantic truth.
Does this contain a recognized watermark?
A detector can support presence of a particular signal or recovery of a payload under its tested conditions. Validate the expected payload and any cryptographic authentication separately. A failed check may reflect unsupported marks, absent marking or damaged signals.
Who signed which claims about this asset?
Check manifest binding, signature, identity and trust policy separately. A valid signature authenticates a signer's assertion and protected content relationship. The signer may be mistaken or dishonest, and a legitimate asset may have lost its manifest.
What capture pipeline was attested?
Device- or app-backed capture can narrow the trusted path from capture to receipt. Its strength depends on hardware, app integrity, key custody and enrollment. Recording a staged scene or another screen remains a separate issue.
Did these bytes or this digest exist by a particular time?
A trusted timestamp or verifiable log can support existence by a time, subject to its trust model. A locally written timestamp is only a local assertion. Neither is automatically the time the depicted event happened.
Do observed traces resemble manipulation or synthesis?
Statistical and learned detectors return evidence about a defined task and population. Scores need calibration, domain checks and error estimates. A clean result does not prove camera origin; an anomalous result does not by itself prove deception.
Does the claimed story fit independent evidence?
Source history, other views, recorded events and expert analysis can corroborate context. A genuine file can carry a false caption, and synthetic media can accurately illustrate a real event. Content origin and factual claims require separate review.
A practical order for LegitPix
Recommended integration order, aligned with local receipt generation and verification. These are proposed stages; this reference library does not add features to the LegitPix app.
1 Preserve and identify
Keep the received file unchanged. Record SHA-256, byte length, content type, receipt schema/version and observation time locally. If normalized pixels or decoded audio are also hashed, label that derivative algorithm and decoding rules separately. A receipt should say exactly what was compared.
2 Inspect claims
Read metadata and C2PA manifests without treating their presence as validation. Separate absent, unsupported, malformed, cryptographically invalid, cryptographically valid but untrusted, and trusted under a named policy. Capture trust-list freshness and revocation limitations, especially offline.
3 Retrieve relationships
Add perceptual matching only against an authorized reference set. Show the matching reference, algorithm/version, distance and threshold, then require a source comparison before making stronger claims. Keep exact-byte equality and near-duplicate similarity separate.
4 Check specific marks
Use named watermark adapters with clear provider/model coverage, key or payload requirements and measured operating thresholds. Prefer local implementations when practical. Ask for permission before sending personal media to any external provider and disclose what leaves the device.
5 Analyze cautiously
Add passive image, audio and video detectors as qualified analysis after representative evaluation. Show the detector, model hash/version, input scope, score type, calibration population and limitations. Keep unsupported, not tested and inconclusive outcomes available.
6 Strengthen future capture
Evaluate signed receipts, independent timestamps and hardware-backed capture as separate extensions. Define exactly which trust boundary each extension strengthens. Do not relabel an existing imported photograph as sensor-attested.
Keep result labels specific.
- Exact-byte match to receipt [receipt ID].
- Signed provenance validates under [trust policy]; assertions shown separately.
- Possible source match to [reference] at [algorithm/version/threshold].
- Recognized [provider/method] watermark detected; payload/authentication status recorded.
- Forensic indicator of [defined task]; score, scope and limitations available.
- No supported evidence found, unsupported input, or inconclusive.
How to read the catalog
Each record combines the mechanism, supported claim, proof limits, implementation/access, offline feasibility, cost dependencies, code/model/data license notes, version/date, transformation behavior, error caveats, privacy, suggested LegitPix role, uncertainties and sources.
Evidence types are source categories, not quality scores. A standard defines rules. Research reports a particular study. Software repositories document an implementation. Vendor documentation describes a service. Publication, available code and production readiness are different milestones.
Offline-capable does not mean browser-ready. A downloadable model may need native libraries, large weights or a GPU. An external source URL opens only when you click it. Search and filters in this file run locally and send no media or queries anywhere.
Method counts are not an exhaustive census. Some records are families, others are named implementations or standards. Read the family and version fields before comparing entries.
For an advocacy and information site
Proposed editorial structure, reflecting the direction shared on 6 October 2026. No public site was built or published for this library. These are content recommendations, not a statement of the owner’s policy positions.
Suggested public topic structure
Start here
Explain what file integrity, provenance, watermarks, source matches and detector results can establish. Make the difference between media origin and the truth of a claim easy to understand.
Methods and standards
Offer a plain-language summary first, with technical records, source dates, implementation availability and limitations one level deeper. Let visitors compare claims rather than a single trust score.
Practical guides
Help creators preserve originals and provenance, and help readers inspect evidence without treating missing metadata or a detector score as a verdict. Explain privacy choices before external checks.
Evidence in practice
Use rights-cleared examples and clearly labeled controlled demonstrations to show how ordinary editing and reposting affect different signals. Separate measured outcomes from hypotheses and product claims.
The case for understandable evidence
An advocacy section could explore clearer disclosure, interpretable verification results and responsible handling of uncertainty. Choose specific positions and calls to action separately; the catalog does not settle them.
Sources and corrections
Show review dates, versions, original sources, conflicts of interest where known, and a visible correction history. Separate research notes, tested guides and vendor claims.
Suggested editorial principles
- State the claim before the method; use the same terms consistently.
- Make limitations visible beside benefits, especially false-positive risks and missing-evidence cases.
- Label proposed, experimental, independently tested and provider-reported material distinctly.
- Avoid universal real/fake scores, guarantees of ownership, or presenting an AI score as proof of dishonesty.
- Keep privacy, consent, accessibility and rights-cleared examples in the review checklist.
- Link primary evidence, date updates and explain material corrections.
LegitPix is intended as a free advocacy and information resource. Signing services, accounts, plugins, sponsorships and other implementation or funding ideas remain proposals unless separately verified as operating.
Existing public education resources
- CAI public explanation and FAQ · Reader-facing companion explaining provenance and why credentials do not decide whether media is real. Use for terminology and evidence literacy, not independent proof of product efficacy. · checked 2026-10-12
- CAI media literacy curriculum · Resource for adapting visual/media-literacy lessons to public education. Official page describes school-age curricula and subject-specific lessons. Materials' reuse license was not verified; link rather than copy or republish them. · checked 2026-10-12
- Content Credentials Foundations · Existing implementer course covering signing, display and advanced topics. Listed as an educational reference, not as a newly launched tool or independent validation. · checked 2026-10-12
A path toward a first product
A practical first-product decision path. This is guidance for choosing and validating a product, not authorization to implement or publish it. Recommendations below are architectural judgments from the evidence boundaries, not head-to-head performance findings.
Start with a public evidence-literacy repository paired with a narrowly scoped, local-first photo receipt and evidence inspector. Preserve the existing distinction between byte consistency and image truth. Add methods only when they answer a clearly defined user question.
Recommended method combinations
First combination Exact receipts plus explainable provenance
Use whole-file SHA-256 receipts as the integrity foundation. Surface metadata as claims. Add C2PA inspection/validation only after its runtime, trust-policy and offline behavior pass tests. Export a readable evidence record with the compared file, algorithm/version, source of time, validation state and limitations.
Second combination Reference matching where a reference exists
If users need to recognize changed copies, evaluate a perceptual matcher against an authorized collection. Return candidate source links and distances alongside the exact-hash result. Do not interpret a similarity match as independent verification of the scene.
Third combination Specific watermark checks
Add a named watermark detector when relevant producers or workflows actually use it. Record provider/model coverage, expected payload or key, and validation status. Keep hosted checks opt-in with a clear explanation of what media is transmitted.
Later combination Forensic analysis and stronger capture
Treat passive AI/tamper detectors as additional indicators only after domain-specific calibration. Evaluate capture signing and external timestamps separately when users need evidence created at capture time or independently witnessed existence. These solve different problems.
Decisions and gates before release
Choose one first user job
Decide whether the first tool primarily helps creators issue receipts or readers inspect received files. Use that choice to define the minimal flow, language and test collection.
Pass the evidence and failure-state gate
Demonstrate exact-byte comparisons, invalid receipts, absent manifests, unsupported formats and malformed inputs. Keep signature validity, signer trust, identity and source claims separate. Never turn missing evidence into a negative authenticity verdict.
Pass the privacy and rights gate
Verify local processing, data minimization, optional external transmission, code/weights/data licenses and model provenance. Do not add a vendor SDK solely because its documentation is public.
Pass the deployment evaluation gate
Benchmark the chosen combination on rights-cleared, representative inputs and ordinary delivery transformations. Publish the tested version, scope and known limitations. Define a tolerable false-positive rate and inconclusive policy before showing any detector verdict.
Release a narrow claim and collect evidence
Launch only the behavior that passed the gates. Keep the catalog separate from the implemented-method list, and use a correction/version history. Gather product feedback without silently collecting users’ media.
Method catalog
Search names, technologies, claims, license notes or limitations. Combine filters to narrow the reference. Sources and implementation details expand within each record.
90 of 90 records
No records match these filters. Try a broader term or reset the filters.
ai detectionfd-aasist
Audio anti-spoofing classifier (AASIST / ASVspoof baselines)
Learn spectro-temporal graph-attention representations to distinguish bona fide speech from synthetic/converted speech in defined tasks.
What it supports
A speech anti-spoofing score within the evaluated recording/channel/generator domain.
What it does not prove
Speaker identity, truthful words, consent, or universal synthetic-audio detection.
Implementation, limitations and sources
- Implementation and access
- Official AASIST/AASIST-L code and pretrained models; separate ASVspoof5 baseline.
- Offline feasibility
- Yes after model assets.
- Cost and dependencies
- Audio preprocessing and model inference; representative speech evaluation.
- Licenses and rights
- Original AASIST MIT; ASVspoof5 assets and speech datasets need separate license/consent review.
- Version or date
- ICASSP 2022; ASVspoof5 baseline 2024.
- Behavior under transformations
- Codecs, resampling, microphones, language and playback/re-recording create channel shifts.
- False positives and false negatives
- A low challenge EER is not expected real-world error. Threshold, spoof prevalence and attack coverage change precision.
- Privacy and data rights
- Speech contains personal/biometric information; local processing and limited retention.
- Potential LegitPix role
- Optional speech-specific evidence, never a voice-authentication verdict.
- Uncertainty and open checks
- Benchmark task definitions and trained checkpoint must accompany results.
Primary sources
ai detectionfd-cnn-detector
CNN-generated image classifier (CNNDetection baseline)
Supervised classifier learns synthesis artifacts, with blur/JPEG augmentation used to improve transfer across generation methods.
What it supports
A model score for resemblance to its training/evaluation classes.
What it does not prove
AI generation as fact, exact generator attribution, or real-world authenticity.
Implementation, limitations and sources
- Implementation and access
- Author PyTorch code, training/evaluation scripts and released weights.
- Offline feasibility
- Yes after assets are downloaded.
- Cost and dependencies
- Inference compute and representative evaluation data.
- Licenses and rights
- Repository CC BY-NC-SA 4.0; data/dependencies require separate review.
- Version or date
- CVPR 2020; repository notes dataset-link repair July 2024.
- Behavior under transformations
- Augmentation covers some blur/recompression; modern pipelines may differ substantially.
- False positives and false negatives
- Author demo explicitly calls its output uncalibrated. Old benchmark performance cannot establish current-generator accuracy.
- Privacy and data rights
- Do not accuse creators or make consequential decisions from a classifier output.
- Potential LegitPix role
- Historical benchmark/control, not a production authenticity gate.
- Uncertainty and open checks
- No modern target-domain evaluation performed.
Primary sources
ai detectionfd-dire
Diffusion reconstruction-error detection (DIRE)
Compare the input with a pretrained diffusion model's reconstruction, then classify the reconstruction-error representation.
What it supports
An input's compatibility with learned diffusion reconstruction/detection behavior.
What it does not prove
All AI generation, exact generator identity, untouched capture or event truth.
Implementation, limitations and sources
- Implementation and access
- DIRE code, diffusion reconstruction assets and detector checkpoints.
- Offline feasibility
- In principle after all assets; GPU-oriented and heavier than a simple classifier.
- Cost and dependencies
- Diffusion reconstruction latency/memory plus classifier.
- Licenses and rights
- No clear top-level license located in reviewed official repository; weights/data rights unverified.
- Version or date
- 2023-03-16 preprint; ICCV 2023.
- Behavior under transformations
- Reconstruction behavior depends on chosen diffusion model and input transformations.
- False positives and false negatives
- Reported cross-diffusion transfer is conditional on tested generators/data; edited photos and new models need separate assessment.
- Privacy and data rights
- Local processing possible; do not upload private images to a public demo without approval.
- Potential LegitPix role
- Research/reference-only until rights and operational calibration are established.
- Uncertainty and open checks
- Public source availability is not permission for commercial redistribution.
Primary sources
ai detectionfd-probe
Failure-focused detector training (PROBE)
Use difficult generative variations to refine detectors and improve transfer to unseen generation conditions.
What it supports
Research evidence that targeted variation coverage can improve a detector on evaluated benchmarks.
What it does not prove
Universal detection, robustness to all new generators, or calibrated authenticity probability.
Implementation, limitations and sources
- Implementation and access
- PROBE-AIGI-Detection; ResNet50 and DINOv2 detector checkpoints linked on ModelScope.
- Offline feasibility
- Potentially after assets; end-to-end training release completeness unverified.
- Cost and dependencies
- Inference backbone; training needs generator access and additional compute.
- Licenses and rights
- No top-level code license found in reviewed repository; checkpoint/data terms unverified.
- Version or date
- ICML 2026; PMLR volume 306 published 2026-09-29.
- Behavior under transformations
- Research explores generation variation; ordinary distribution shifts remain a validation requirement.
- False positives and false negatives
- More difficult examples can improve benchmark transfer without establishing real-world false-positive rates.
- Privacy and data rights
- Training/evaluation data permissions and source integrity remain necessary.
- Potential LegitPix role
- Current research watchlist and evaluation-design input.
- Uncertainty and open checks
- README has unchecked release TODOs alongside evaluation/checkpoint links; do not claim a complete training release.
Primary sources
ai detectionfd-effort
Foundation-model subspace adaptation (Effort)
Preserve principal pretrained feature components and adapt an orthogonal residual subspace to reduce detector overfitting.
What it supports
A trained model's anomaly/classification signal for the chosen face or general-image domain.
What it does not prove
Domain-independent accuracy, provenance, generator identity or an untouched image.
Implementation, limitations and sources
- Implementation and access
- Effort-AIGI-Detection; face-specific and general-image checkpoints use different training data/preprocessing.
- Offline feasibility
- Yes after checkpoint/dependency acquisition.
- Cost and dependencies
- CLIP-scale inference, preprocessing and evaluation.
- Licenses and rights
- Top-level universal grant not located; bundled DeepfakeBench is CC BY-NC 4.0. Check each code subtree and checkpoint.
- Version or date
- ICML 2025, PMLR 267.
- Behavior under transformations
- Foundation knowledge may improve transfer; face crops and general images require the appropriate model.
- False positives and false negatives
- Better benchmark transfer does not calibrate deployment scores. Wrong checkpoint/preprocessing can invalidate interpretation.
- Privacy and data rights
- Face processing is sensitive; avoid identity or intent conclusions.
- Potential LegitPix role
- Research candidate for independently measured detector evidence.
- Uncertainty and open checks
- No weight license audit or runtime reproduction performed.
Primary sources
ai detectionfd-universal-fake
Frozen-foundation-feature detector (UniversalFakeDetect)
Use a frozen pretrained CLIP visual representation with a lightweight supervised detector to improve cross-generator transfer.
What it supports
Model-specific synthetic-image evidence within an evaluated domain.
What it does not prove
Universality, truthful camera origin, exact model attribution, or calibrated probability.
Implementation, limitations and sources
- Implementation and access
- UniversalFakeDetect PyTorch/CLIP; official repository moved to WisconsinAIVision.
- Offline feasibility
- Yes once detector and CLIP weights are present.
- Cost and dependencies
- Foundation-model inference and target-domain validation.
- Licenses and rights
- Repository MIT; upstream weights and dataset rights must still be checked.
- Version or date
- CVPR 2023; official main observed 2026-10-06.
- Behavior under transformations
- Transfer evaluated across different generators; preprocessing and distribution shifts remain material.
- False positives and false negatives
- A default 0.5 threshold is not a calibrated probability. Semantic/domain shortcuts and class imbalance can affect results.
- Privacy and data rights
- Prefer local inference and retain scores as tentative evidence.
- Potential LegitPix role
- Candidate baseline for calibration experiments, with model/version recorded.
- Uncertainty and open checks
- Title's 'universal' describes a research aim, not a guarantee.
Primary sources
ai detectionfd-gend
Parameter-efficient cross-benchmark detection (GenD)
Adapt a pretrained visual encoder through LayerNorm tuning and normalized metric learning; evaluate cross-dataset deepfake transfer.
What it supports
Model-specific facial deepfake evidence under its preprocessing and training domain.
What it does not prove
Authenticity across all video types, calibration, or generator/source attribution.
Implementation, limitations and sources
- Implementation and access
- GenD PyTorch/Transformers, face preprocessing and model-specific backbone.
- Offline feasibility
- Yes after model assets; repository demonstrates local inference/UI.
- Cost and dependencies
- Backbone compute and face detection; GPU useful.
- Licenses and rights
- Repository MIT; each backbone/checkpoint/data license must be reviewed separately.
- Version or date
- WACV 2026; code observed 2026-10-06.
- Behavior under transformations
- Must validate crops, compression, non-face video and modern generator changes.
- False positives and false negatives
- Authors emphasize paired real/fake sources to reduce shortcut learning. Cross-benchmark AUROC is not a deployed probability.
- Privacy and data rights
- Limit face data retention and never infer personal identity from detector results.
- Potential LegitPix role
- Current research candidate with explicit model and calibration profile.
- Uncertainty and open checks
- No checkpoint download or independent reproduction; backbone variants need separate review.
Primary sources
ai detectionfd-tall
Spatiotemporal facial deepfake detection (TALL)
Arrange sampled video frames into a thumbnail layout so a visual model can learn spatial/temporal inconsistencies.
What it supports
A model score for facial manipulation patterns in the sampled clip.
What it does not prove
Untampered complete video, truthful event context, intact audio or all-generator coverage.
Implementation, limitations and sources
- Implementation and access
- TALL4Deepfake PyTorch/Swin-based implementation and frame-sequence preparation.
- Offline feasibility
- Yes with the necessary checkpoint/environment.
- Cost and dependencies
- Video decoding, face crops, frame sampling and model inference.
- Licenses and rights
- Official repository MIT; checkpoints/backbones/datasets separately reviewable.
- Version or date
- ICCV 2023; TALL++ IJCV 2024 announcement.
- Behavior under transformations
- Frame rate, temporal sampling, face visibility and recoding can affect the evidence.
- False positives and false negatives
- A model can miss unsampled alterations; benign temporal artifacts can cause false alarms. Face-domain training is not general video authentication.
- Privacy and data rights
- Limit face-frame retention and avoid identity conclusions.
- Potential LegitPix role
- Research video-detector candidate, distinct from reference-based temporal matching.
- Uncertainty and open checks
- README still says TALL++ code will be made available; do not infer the extension is fully released.
Primary sources
ai detectionfd-trufor
TruFor manipulation detection and localization
Fuse RGB and learned noise-sensitive features to produce manipulation localization, an image-level score and a reliability map.
What it supports
Model-estimated local manipulation anomalies and associated uncertainty.
What it does not prove
That every clean-looking image is genuine, which editor was used, or what happened in the scene.
Implementation, limitations and sources
- Implementation and access
- Author PyTorch implementation; Docker inference documented.
- Offline feasibility
- Yes after weights and environment are acquired.
- Cost and dependencies
- Typically GPU-friendly inference and substantial model dependencies.
- Licenses and rights
- Informational/nonprofit-only author license; no unrestricted commercial deployment inferred.
- Version or date
- CVPR 2023; training code announced 2025-03-05.
- Behavior under transformations
- Evaluated across conventional and learning-based manipulation types; operational transformations still need testing.
- False positives and false negatives
- Localization/reliability maps are model outputs, not calibrated proof. Domain shift, benign processing and small edits can cause errors.
- Privacy and data rights
- Keep original and derived maps private; maps can be reputationally consequential if overstated.
- Potential LegitPix role
- Opt-in research anomaly maps with uncertainty and expert review.
- Uncertainty and open checks
- Noiseprint++ training release remained a TODO in reviewed README; no runtime test.
Primary sources
audio matchingfd-chromaprint
Chromaprint compact audio fingerprints
Generate compact fingerprints intended for near-identical recordings, duplicates and long-stream monitoring.
What it supports
Recording similarity against a compatible reference index.
What it does not prove
Speaker identity, truthful speech, ownership, or absence of audio synthesis.
Implementation, limitations and sources
- Implementation and access
- acoustid/chromaprint; FFmpeg commonly supplies decoding; FFT backend configurable.
- Offline feasibility
- Yes for fingerprint computation and a separately implemented/local matcher.
- Cost and dependencies
- Local compute; AcoustID service/index is optional and separate.
- Licenses and rights
- Own code MIT, but LICENSE.md says project as a whole LGPL-2.1 because included FFmpeg code; FFTW can impose GPL obligations.
- Version or date
- Official master README/LICENSE observed 2026-10-06.
- Behavior under transformations
- Optimized for near-identical audio; not a general-purpose noisy short-clip recognizer.
- False positives and false negatives
- Library explicitly trades precision/robustness for search speed; silence, short clips and major alterations need special handling.
- Privacy and data rights
- Audio fingerprints still permit linkage to recordings.
- Potential LegitPix role
- Optional recording-match evidence; do not confuse with an audio watermark detector.
- Uncertainty and open checks
- Audit exact source version and compiled dependency licenses before shipping.
Primary sources
audio matchingfd-audfprint
Landmark-based audio fingerprinting (audfprint)
Index spectral landmarks and find temporally consistent matches for noisy excerpts.
What it supports
Candidate recording/segment reuse with an estimated offset.
What it does not prove
What was said is true, who spoke, or whether the source is synthetic.
Implementation, limitations and sources
- Implementation and access
- Dan Ellis audfprint Python package; FFmpeg decoding.
- Offline feasibility
- Yes; local database creation and queries.
- Cost and dependencies
- CPU, reference database size and landmark density.
- Licenses and rights
- MIT code; FFmpeg and Python dependencies separately licensed.
- Version or date
- Official repository observed 2026-10-06; historical README examples are not a current release date.
- Behavior under transformations
- Designed for excerpts and noise; timing/encoding/channel changes still require evaluation.
- False positives and false negatives
- Shared music, jingles and low-information sound can produce ambiguous matches. Distinguish hash collisions from sustained aligned evidence.
- Privacy and data rights
- Audio indexing may expose speech/music membership; keep authorized local references.
- Potential LegitPix role
- Audio-segment relationship matcher with offsets and matched duration.
- Uncertainty and open checks
- No production scalability or target-domain accuracy claim verified.
Primary sources
audio matchingfd-panako
Panako time/pitch-tolerant audio matching
Audio fingerprinting research system with algorithms designed to recognize time-scale and pitch-modified recordings.
What it supports
A candidate recording relationship under wider timing/pitch variation.
What it does not prove
Speaker authenticity, musical ownership, or truthful content.
Implementation, limitations and sources
- Implementation and access
- Java/Gradle, FFmpeg, LMDB and Gaborator-related dependencies.
- Offline feasibility
- Yes after installation; platform-specific dependencies matter.
- Cost and dependencies
- Compute and local index; potentially significant integration/legal review.
- Licenses and rights
- AGPL source; dependency licenses vary. README raises patent considerations; present enforceability not assessed.
- Version or date
- README changelog includes Panako 2.1, May 2022; observed 2026-10-06.
- Behavior under transformations
- Explicit focus on speed, time-stretch and pitch variation.
- False positives and false negatives
- Claims apply to tested material/settings; do not transfer music retrieval performance to all speech/noise.
- Privacy and data rights
- Use permitted reference audio; database access can reveal listening or recording relationships.
- Potential LegitPix role
- Specialized research candidate when simple audio matching is inadequate.
- Uncertainty and open checks
- README contains stale platform statements alongside newer updates; verify current build before adoption.
Primary sources
capturecapture-truepic
Controlled capture and signing: Truepic
Controlled capture with product-specific assurance: native Vision uses device-key signing and device-integrity checks; browser and embedded-WebView approaches use server-side signing. Separate Lens server products can sign existing files.
What it supports
Evidence about an explicitly identified controlled capture or signing workflow.
What it does not prove
Every Truepic signature originated at capture, or that a scene/caption is factual. Server signing is a separate workflow.
Implementation, limitations and sources
- Implementation and access
- Truepic mobile/browser capture integrations and Lens server CLI/API; verify exact product scope.
- Offline feasibility
- Native Vision supports offline capture. Web/embedded approaches provide pending-upload retry rather than a full offline workflow. Lens server enrollment requires connectivity; timestamping may require it.
- Cost and dependencies
- Enterprise integration, CA/service access and negotiated price; no free general detector confirmed.
- Licenses and rights
- Proprietary contracts; Lens CLI/display library treated as confidential in current MSA.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- C2PA hard-binding rules apply; capture checks and downstream edit history remain distinct.
- False positives and false negatives
- Browser/embedded approaches do not provide native device attestation or root/jailbreak checks. Vendor assurance claims require exact product, platform and threat model; independent universal error rates unavailable.
- Privacy and data rights
- Capture may involve location/device attestation; server Lens MSA describes usage telemetry. Minimize and review contract.
- Potential LegitPix role
- Optional capture-origin adapter; label capture evidence separately from post-hoc signing.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
capturecapture-leica
Hardware-backed capture signing: Leica M11-P example
Camera signs captured images using protected hardware and provisioned certificates when Content Credentials is enabled.
What it supports
A valid credential attributes captured-file evidence to the camera’s signing implementation.
What it does not prove
The scene was unstaged, correctly captioned, or not a photograph of another image/display.
Implementation, limitations and sources
- Implementation and access
- Leica M11-P integrated signing; independent C2PA viewers can inspect supported outputs.
- Offline feasibility
- Camera capture signing and local signature verification possible; trust/status freshness remains separate.
- Cost and dependencies
- Specific camera hardware and lifecycle support; current purchase price not evaluated.
- Licenses and rights
- Proprietary firmware/hardware; open C2PA format.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Direct signatures bind original output; derivatives need preserved provenance or new compliant signing.
- False positives and false negatives
- Validate actual certificate/firmware and signed fields; camera claims do not make arbitrary EXIF true.
- Privacy and data rights
- Device/creator metadata can identify a photographer; inspect what is published.
- Potential LegitPix role
- Higher-assurance capture-origin evidence when present; never prerequisite for a legitimate image.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
capturecapture-sony
Sony Camera Authenticity Solution
Hardware signing combines C2PA with proprietary depth/recapture analysis, modification analysis and server-time evidence.
What it supports
Valid camera-signing evidence plus separately reported vendor analysis under supported capture conditions.
What it does not prove
Scene/event truth or harmless intent. A real 3D object can still be staged or misleading.
Implementation, limitations and sources
- Implementation and access
- 2026.2 service; compatible still/video cameras; Sony Self-Checker and validation site. Image Validation API supports still images only; pre-check SDK supports stills and video. Camera Verify, Stamping Tool and SignatureViewer.js remain Beta.
- Offline feasibility
- Basic local signature/pre-check possible; advanced Sony validation is subscription-dependent.
- Cost and dependencies
- Paid digital-signature upgrade and validation service; default certificate is free but trust-list status must be checked.
- Licenses and rights
- Proprietary hardware/software/service; C2PA format open.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Compatible editing can preserve history; specific video-format/bitrate restrictions apply.
- False positives and false negatives
- Do not equate generic C2PA validation with Sony’s proprietary 3D or modification results; independent error rates not established.
- Privacy and data rights
- Advanced verification may upload media/depth metadata; authorization required.
- Potential LegitPix role
- Import granular external evidence if available; avoid exclusive vendor dependency.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
exact integrityfd-decoded-hash
Canonical decoded-content and per-frame hashing
Hash decoded pixels or audio samples under an explicitly fixed decoding/canonicalization profile; FFmpeg supplies hash, framehash and streamhash muxers.
What it supports
Equality of selected decoded data under the recorded profile, including frame-level comparison.
What it does not prove
Identical container bytes, all metadata, temporal context, or authenticity.
Implementation, limitations and sources
- Implementation and access
- FFmpeg; pin decoder/build, streams, pixel/sample format, dimensions, color handling and timing policy.
- Offline feasibility
- Yes.
- Cost and dependencies
- Decoding cost; codec support and build variability.
- Licenses and rights
- FFmpeg usually LGPL-2.1-or-later; enabled components can make a build GPL or otherwise restricted.
- Version or date
- Rolling FFmpeg documentation observed 2026-10-06.
- Behavior under transformations
- Container-only changes may leave decoded content equal; resizing, color conversion or lossy recoding usually will not.
- False positives and false negatives
- Different decoders or color/sample conversions may produce different bytes. Equal sampled frames do not cover omitted frames.
- Privacy and data rights
- Local processing preferred; preserve raw original separately.
- Potential LegitPix role
- Optional diagnostic relationship below whole-file integrity.
- Uncertainty and open checks
- No universal decoded-media canonical form is specified here; not runtime-tested.
Primary sources
exact integrityfd-sha256
Cryptographic whole-file hashing (SHA-256)
Hash the exact received bytes and compare with a trusted recorded digest.
What it supports
Extremely strong byte-identity/change evidence under the hash's security assumptions.
What it does not prove
Capture origin, author, time, consent, truth, or whether the reference was already manipulated.
Implementation, limitations and sources
- Implementation and access
- Standard cryptographic libraries; retain algorithm and digest with receipt.
- Offline feasibility
- Yes; no media upload required.
- Cost and dependencies
- Negligible CPU; trust in the reference record is separate.
- Licenses and rights
- Algorithm standard; chosen implementation has its own license.
- Version or date
- FIPS 180-4, August 2015; NIST revision planned.
- Behavior under transformations
- Any byte change normally changes the digest, including harmless metadata edits.
- False positives and false negatives
- A matching hash without an authenticated reference says only that two inputs match. Hashing alone is not signing.
- Privacy and data rights
- Hashes can link records and permit known-file membership tests; do not treat them as anonymized.
- Potential LegitPix role
- Core byte-consistency receipt; bind claims to the original file digest.
- Uncertainty and open checks
- Does not establish truth before the first trusted observation.
Primary sources
hosted ai detectionfd-hive-detection
Commercial AI-generated and face-deepfake detection (Hive)
Hosted visual API combines generated-image/video classification with face-deepfake classification; a separate audio model analyzes synthetic audio.
What it supports
Provider-scoped model scores and tentative generator-family labels.
What it does not prove
A trusted generator attribution, video/audio integrity, absence of manipulation or rights ownership.
Implementation, limitations and sources
- Implementation and access
- Authenticated API; record visual, face and audio outputs separately. Visual video results include sampled-frame decisions.
- Offline feasibility
- Hosted service; no public offline weights verified. Private deployment terms unconfirmed.
- Cost and dependencies
- Usage-based service; image, video-frame and audio processing have different units. Current contract and total workload cost not verified.
- Licenses and rights
- Proprietary models/service terms; not an open-source detector.
- Version or date
- Official image/video, audio and FAQ documentation accessed 2026-10-06.
- Behavior under transformations
- Frame sampling, compression, face visibility and audio channel changes affect applicability.
- False positives and false negatives
- Vendor recommends use-case threshold refinement. Maximum face score and per-frame aggregation can amplify errors across many faces/frames.
- Privacy and data rights
- FAQ states default customer-data retention of 14 days, adjustable on request; zero retention must be established for the actual project.
- Potential LegitPix role
- Optional external detector evidence; distinguish neural scores from any separately returned C2PA metadata.
- Uncertainty and open checks
- No customer account, private deployment, model-version guarantee or independent performance test performed.
Primary sources
hosted ai detectionfd-sightengine-detection
Commercial image/video, deepfake and audio detection (Sightengine)
Hosted models offer generated-image/video, face-deepfake, music and speech analysis, with global and some generator-specific scores.
What it supports
A provider-produced detector observation for the selected model and input.
What it does not prove
True provenance, exact generating system, genuine voice identity or trustworthy event content.
Implementation, limitations and sources
- Implementation and access
- Authenticated image/video/audio APIs; preserve modality-specific scores and generator-label schema. October 2026 expands named image/video classes previously returned as other.
- Offline feasibility
- Hosted API; no public offline model distribution or private deployment verified.
- Cost and dependencies
- Plans and operation/model-based billing; speech docs restrict access to enterprise users/partners, despite broader paid-plan FAQ wording.
- Licenses and rights
- Proprietary subscription/model terms; checkpoint licenses not available.
- Version or date
- October 2026 changelog adds image/video generator classes previously grouped as other; June 2026 speech introduction. Sources reviewed 2026-10-12.
- Behavior under transformations
- Frame selection, image rescaling and audio channel changes need separate validation.
- False positives and false negatives
- Performance and generator labels remain provider claims; scores across modalities/models are not automatically comparable or calibrated.
- Privacy and data rights
- Uploads or fetched media URLs expose content to the service. Confirm subscription retention/subprocessors; privacy policy allows agreement-dependent retention.
- Potential LegitPix role
- Optional gated vendor evidence after scope, privacy and threshold evaluation.
- Uncertainty and open checks
- April 2026 guide still describes speech as in development; current June changelog/docs used. No account, media upload or contractual audit.
Primary sources
hosted ai detectionfd-reality-defender
Commercial multimodal deepfake detection (Reality Defender RealAPI)
Proprietary models return ensemble manipulation scores and supporting indicators through an authenticated upload/API workflow.
What it supports
A provider-generated analysis result for the submitted media and model version.
What it does not prove
Authenticity, identity, event truth or a calibrated real-world probability merely because the provider calls its score a probability.
Implementation, limitations and sources
- Implementation and access
- REST/SDK file upload then polling or webhook; account/API key required.
- Offline feasibility
- Hosted API requires network; enterprise page advertises on-premises, private-cloud, containerized and air-gapped laptop deployments, subject to contract.
- Cost and dependencies
- Free tier lists image/audio; video is included in paid tiers. Enterprise pricing/deployment quoted separately; billing details not audited.
- Licenses and rights
- Proprietary service/models; SDK availability does not license model weights.
- Version or date
- Current RealAPI product and quickstart pages accessed 2026-10-06.
- Behavior under transformations
- Compression, media quality and unseen generators require target-specific testing.
- False positives and false negatives
- Treat claimed performance as vendor evidence. Preserve modality/model details and independently calibrate thresholds.
- Privacy and data rights
- Hosted uploads leave the device; organization admins can view users' uploads. Verify retention, locality and contractual processing before use.
- Potential LegitPix role
- Optional commercial adapter after data-sharing approval and validation; never collapse its result into verified provenance.
- Uncertainty and open checks
- Older FAQ access language conflicts with current self-serve product page; current product page used. No account, upload, quote or private deployment tested.
Primary sources
metadataprov-ordinary-metadata
EXIF/XMP/IPTC metadata and Digital Source Type
Structured fields describe camera, creator, location, software, rights and declared source type; IPTC adds AI system/prompt fields.
What it supports
The file contains these declarations; stronger attribution requires authenticated inclusion in a signed assertion.
What it does not prove
Values are true, the named camera produced the pixels, absence of AI, or complete editing history.
Implementation, limitations and sources
- Implementation and access
- Common metadata parsers/editors; IPTC 2025.1 current published standard.
- Offline feasibility
- Yes; conventional parsing.
- Cost and dependencies
- Open tools available; no mandatory hosted service.
- Licenses and rights
- IPTC specification CC BY 4.0; format/tool terms separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Platforms/editors may alter or discard fields; changing covered metadata can invalidate C2PA.
- False positives and false negatives
- Treat unsigned source-type and software fields as claims, not positive or negative AI detection.
- Privacy and data rights
- GPS, names and prompts may be sensitive; do not expose all metadata by default.
- Potential LegitPix role
- Context/explanation layer labelled “declared metadata,” separate from signed evidence.
- Uncertainty and open checks
- This record chiefly covers photo/video metadata standards. Audio container-specific tags and applications need separate treatment.
Primary sources
multi level identifierfd-iscc
ISCC multi-component content identifier (ISO 24138)
Build a typed composite identifier from metadata, perceptual/content, data-similarity and exact-instance components.
What it supports
Component-specific exact or similarity relationships for a digital asset.
What it does not prove
Ownership, registered authorship, trusted capture or truth of content.
Implementation, limitations and sources
- Implementation and access
- iscc-core for core algorithms; higher-level extraction is delegated to iscc-sdk.
- Offline feasibility
- Core generation yes; format extraction/dependencies and any registry service are separate.
- Cost and dependencies
- Local computation; optional search/index/registry infrastructure.
- Licenses and rights
- iscc-core Apache-2.0; full ISO text and SDK dependencies have separate terms.
- Version or date
- ISO 24138:2024, edition 1, May 2024.
- Behavior under transformations
- Similarity preservation depends on selected unit; instance checksum remains exact.
- False positives and false negatives
- Do not interpret a soft component as a cryptographic commitment. Registration and trustworthy declarations are separate.
- Privacy and data rights
- Identifiers enable cross-system linkage and membership tests.
- Potential LegitPix role
- Interoperable optional asset identifier alongside byte receipts and explicit evidence states.
- Uncertainty and open checks
- Semantic-Code is reserved, not algorithmically standardized by the 2024 edition; experimental extensions must be labeled.
Primary sources
multimodal forensicsfd-avff
Audio-visual correspondence and fusion (AVFF representative)
Learn correspondences between sound and visual motion, then combine modalities for video deepfake classification.
What it supports
Model-estimated audiovisual inconsistency or synthetic-content evidence.
What it does not prove
That synchronized media is genuine, that a dubbed video is deceptive, or that audio/video share a trusted source.
Implementation, limitations and sources
- Implementation and access
- AVFF paper describes self-supervised correspondence learning followed by supervised classification.
- Offline feasibility
- Conceptually possible; official complete code/weights not established in this review.
- Cost and dependencies
- Two-modality models, synchronization and sufficiently visible speaking subjects.
- Licenses and rights
- Publication access is not a software/model license; third-party reimplementations differ.
- Version or date
- CVPR 2024.
- Behavior under transformations
- Dubbing, translation, latency, edits and audiovisual offsets can change consistency without deception.
- False positives and false negatives
- Matching lip motion can be synthesized; off-screen speech or occlusion can invalidate assumptions. Dataset-specific scores cannot be generalized.
- Privacy and data rights
- Combined face and voice analysis is sensitive; do not infer identity.
- Potential LegitPix role
- Reference/research taxonomy for corroborating modalities, not a deployable default.
- Uncertainty and open checks
- No official implementation/license or operational calibration verified.
Primary sources
passive forensicsforensic-audio-integrity
Audio recording and edit forensics
Compare waveform, spectrogram, encoding structure, bit depth, channel behavior and recording-device exemplars against a specific production claim.
What it supports
Consistency or inconsistency with an alleged recording process, including some edits or pause/restart events.
What it does not prove
Complete absence of editing, speaker identity, intent or truth of recorded statements.
Implementation, limitations and sources
- Implementation and access
- Expert workflow combining file and signal examination; preserve the original encoded file.
- Offline feasibility
- Local examination is possible; exemplars, specialist tools or reference data may be needed.
- Cost and dependencies
- Analyst time, tools and reference data; no prices verified.
- Licenses and rights
- Guidance is not a software license. Tool and reference-data rights must be checked separately.
- Version or date
- SWGDE 15-A-001 v1.3, 20 September 2018; still listed as current on 6 October 2026.
- Behavior under transformations
- Re-encoding can discard format evidence and change traces; decoded audio can retain some signal-level evidence.
- False positives and false negatives
- Ordinary device operations can resemble edits; conclusions require validated methods and controlled exemplars.
- Privacy and data rights
- Recordings and exemplars can contain private conversations and third-party voices.
- Potential LegitPix role
- Explain specialist evidence and support escalation beyond automated scores; not a first-release verdict engine.
- Uncertainty and open checks
- No forensic examination or tool benchmark performed. Guidance is not product certification or endorsement.
Primary sources
passive forensicsfd-cfa
CFA/demosaicing consistency analysis
Estimate periodic color-interpolation traces, such as Bayer mosaic alignment, and flag locally inconsistent patterns.
What it supports
An image region is inconsistent with a particular assumed sampling/demosaicing pattern.
What it does not prove
Specific device identity, universal editing detection or scene truth.
Implementation, limitations and sources
- Implementation and access
- Intermediate Values Algorithm for mosaic analysis.
- Offline feasibility
- Yes; source distribution available.
- Cost and dependencies
- CPU and suitable sufficiently preserved image data.
- Licenses and rights
- IPOL code metadata AGPL-3.0-or-later; version 1.0.
- Version or date
- IPOL publication 2021-10-15.
- Behavior under transformations
- Scaling, rotation, compression and modern multi-frame processing can weaken or alter traces.
- False positives and false negatives
- Bayer assumptions do not cover every sensor or computational-imaging pipeline. Pattern agreement cannot rule out editing.
- Privacy and data rights
- Local analysis preserves confidentiality; retain raw original if available.
- Potential LegitPix role
- Specialist capture-pipeline consistency evidence.
- Uncertainty and open checks
- Applicability should be checked before interpreting any heatmap.
Primary sources
passive forensicsforensic-enf
Electric network frequency evidence in recordings
Analyze time-varying residual mains hum and compare it with suitable historical grid references when a usable signal exists.
What it supports
Qualified evidence about timing or continuity under validated extraction and comparison conditions.
What it does not prove
Guaranteed recording date, precise location, authenticity of speech or absence of editing.
Implementation, limitations and sources
- Implementation and access
- Specialist ENF extraction and database comparison; no turnkey implementation assessed.
- Offline feasibility
- Local examination is possible; exemplars, specialist tools or reference data may be needed.
- Cost and dependencies
- Analyst time, tools and reference data; no prices verified.
- Licenses and rights
- Guidance is not a software license. Tool and reference-data rights must be checked separately.
- Version or date
- SWGDE 09-A-001 v1.2, 6 February 2014; current listing checked 6 October 2026.
- Behavior under transformations
- Noise, duplication, processing and recording conditions can obscure or confound the signal.
- False positives and false negatives
- Hum may be absent. Database coverage, cyclic patterns and multiple signals complicate matching; consider no-reference outcomes.
- Privacy and data rights
- Inferred recording time or grid association can reveal context; protect reference and case data.
- Potential LegitPix role
- Explain specialist evidence and support escalation beyond automated scores; not a first-release verdict engine.
- Uncertainty and open checks
- No forensic examination or tool benchmark performed. Guidance is not product certification or endorsement.
Primary sources
passive forensicsfd-ela
Error Level Analysis (ELA) visualization
Compare an image with a recompressed copy and visualize differences.
What it supports
Only that regions respond differently to that chosen recompression procedure.
What it does not prove
Forgery, AI generation, edit count or authentic untouched regions.
Implementation, limitations and sources
- Implementation and access
- Forensically browser-local ELA; author explicitly warns results can mislead.
- Offline feasibility
- Forensically says yes in modern browsers after assets are cached; images remain local.
- Cost and dependencies
- Free web tool and local browser compute.
- Licenses and rights
- Free use documented; redistribution license for the complete app not verified.
- Version or date
- Official tool changelog includes maintenance update 2025-03-29.
- Behavior under transformations
- JPEG quality, edges, textures, previous exports and re-encoding strongly change the visualization.
- False positives and false negatives
- Brightness/heatmap contrast is not tampering probability. Natural high-detail regions can look suspicious; manipulations can look uniform.
- Privacy and data rights
- Provider documents no image uploads; verify behavior of any deployment or fork.
- Potential LegitPix role
- Expert inspection overlay only; never drive a green/red authenticity badge.
- Uncertainty and open checks
- No app runtime/network inspection performed; offline/privacy behavior is author-reported.
Primary sources
passive forensicsfd-metadata
EXIF/XMP/IPTC/container metadata and thumbnail consistency
Read metadata, embedded previews, software tags, timestamps and file structure; compare claims for consistency.
What it supports
What the file presently declares, plus inconsistencies that merit investigation.
What it does not prove
That a timestamp, camera identifier, location, author or software field is truthful.
Implementation, limitations and sources
- Implementation and access
- Read-only ExifTool extraction; record raw and interpreted values, duplicate fields and parser version.
- Offline feasibility
- Yes.
- Cost and dependencies
- Low local compute; format support/version matters.
- Licenses and rights
- ExifTool under Perl's terms: Artistic License or GPL; dependencies separate.
- Version or date
- Official README observed 2026-10-06; 13.53 documentation verified, not asserted latest.
- Behavior under transformations
- Re-export, messaging platforms and privacy tools may remove or rewrite metadata and thumbnails.
- False positives and false negatives
- Missing tags are ordinary; editing-software tags do not prove deceptive editing. Metadata can be altered independently of pixels.
- Privacy and data rights
- GPS, names and device serials may be sensitive; avoid external extraction uploads.
- Potential LegitPix role
- Descriptive evidence with clear unverified-claim status; signed provenance evaluated separately.
- Uncertainty and open checks
- Do not conflate untrusted metadata with signed content credentials.
Primary sources
passive forensicsfd-copy-move
Internal copy-move detection
Search within an image for duplicated patches/descriptors and geometrically consistent displacement fields.
What it supports
Candidate repeated regions that may result from cloning or object duplication.
What it does not prove
Copying direction, deceptive intent, external splicing, or that naturally repetitive content was edited.
Implementation, limitations and sources
- Implementation and access
- IPOL implementation of dense-field copy-move detection by Ehret, discussing Cozzolino et al.
- Offline feasibility
- Yes; source archive available.
- Cost and dependencies
- Memory/compute rises with image size and dense descriptor processing.
- Licenses and rights
- IPOL source metadata lists GPL-3.0-or-later; code v1.0.
- Version or date
- IPOL 2018; software version 1.0.
- Behavior under transformations
- Methods can cover translations and some rotated/scaled copies; tolerance depends on descriptors/settings.
- False positives and false negatives
- Repeated windows, foliage, textures or generated repeated structures can trigger candidates. Heavy changes and small regions can be missed.
- Privacy and data rights
- Local analysis; retain input and output masks without modifying evidence.
- Potential LegitPix role
- Region-level evidence with side-by-side correspondences, not an automatic forgery verdict.
- Uncertainty and open checks
- Paper page intermittently failed; indexed primary record verified title, method and software license.
Primary sources
passive forensicsfd-jpeg-grid
JPEG grid and compression-history analysis
Recover local JPEG block-grid origins or DCT-statistical traces; compare inconsistent grids/compression evidence.
What it supports
Potential processing-history or region inconsistency compatible with compositing.
What it does not prove
Malicious tampering, how many legitimate edits occurred, or that consistent compression means authenticity.
Implementation, limitations and sources
- Implementation and access
- Local JPEG Grid Detector and ZERO; other double-JPEG tools need their own validation.
- Offline feasibility
- Yes from source.
- Cost and dependencies
- Local CPU and knowledgeable interpretation.
- Licenses and rights
- Both cited IPOL code records AGPL-3.0-or-later; article rights separate.
- Version or date
- Grid detector v2.0, 2020-05-21; ZERO v4.0, 2021-12-16.
- Behavior under transformations
- Recompression, crops and resizing can create, erase or shift traces.
- False positives and false negatives
- Ordinary editing/export can cause multiple-compression evidence. Model-based NFA is not a posterior authenticity probability.
- Privacy and data rights
- Analyze a working copy and retain exact file bytes.
- Potential LegitPix role
- Specialist supporting signal; log detected grids, masks and parameters.
- Uncertainty and open checks
- Not every codec or camera pipeline is covered; no benchmark run.
Primary sources
passive forensicsfd-physical-consistency
Lighting, shadows and scene-geometry consistency
Compare physically modeled illumination or geometry across objects/frames using explicit scene assumptions.
What it supports
Inconsistency with a specified lighting/geometric model; a reason for further investigation.
What it does not prove
Manipulation from visual oddness alone, or authenticity from plausible geometry.
Implementation, limitations and sources
- Implementation and access
- Johnson/Farid illumination research provides point-source and more complex lighting models.
- Offline feasibility
- Yes if suitable images/calibration and analysis tools are available.
- Cost and dependencies
- Expert time, sufficient scene detail and potentially additional views.
- Licenses and rights
- Papers do not provide a blanket software license; choose/verify any implementation separately.
- Version or date
- Primary illumination papers 2005 and 2007.
- Behavior under transformations
- Unknown light sources, reflections, perspective, lens distortion and processing change interpretation.
- False positives and false negatives
- Incorrect physical assumptions can create false alarms; plausible synthetic scenes may satisfy simple models.
- Privacy and data rights
- Scene reconstruction can expose location or private context.
- Potential LegitPix role
- Expert corroboration, with assumptions, alternatives and uncertainty disclosed.
- Uncertainty and open checks
- No ready maintained implementation or measured operating threshold established.
Primary sources
passive forensicsfd-noisesniffer
Noise-level inconsistency (NoiseSniffer)
Use a background stochastic noise model to localize anomalies and report a number-of-false-alarms statistic.
What it supports
Anomalous local noise relative to specified model assumptions.
What it does not prove
That an anomaly is malicious editing, that noise-consistent content is authentic, or the source device.
Implementation, limitations and sources
- Implementation and access
- Python NoiseSniffer and scientific-library dependencies.
- Offline feasibility
- Yes.
- Cost and dependencies
- CPU/memory and image resolution.
- Licenses and rights
- Author repository GPL-3.0; inspect exact license/dependencies before redistribution.
- Version or date
- IWBF 2021; repository observed 2026-10-06.
- Behavior under transformations
- Denoising, tone processing, recompression and heterogeneous image regions influence noise estimates.
- False positives and false negatives
- A model-based false-alarm statistic is not the probability an image is fake; assumptions may fail on computational photography.
- Privacy and data rights
- Local processing available; preserve original and analysis parameters.
- Potential LegitPix role
- Complementary statistical anomaly signal, separated from learned detectors.
- Uncertainty and open checks
- No independent deployment validation; later IPOL refinement exists but full page was unavailable during this pass.
Primary sources
passive forensicsfd-noiseprint
Noiseprint learned camera-model traces
A CNN extracts noise-like processing traces associated with camera models and supports blind localization of inconsistent regions.
What it supports
Processing-trace inconsistency or similarity under the trained model.
What it does not prove
A specific physical camera, full provenance, or an untouched/truthful scene.
Implementation, limitations and sources
- Implementation and access
- Author TensorFlow implementation and models; blind heatmap/extraction tools.
- Offline feasibility
- Yes once assets/dependencies are available.
- Cost and dependencies
- Inference compute; older TensorFlow environment may need isolation.
- Licenses and rights
- Author license restricts use/reproduction/modification to informational and nonprofit purposes.
- Version or date
- 2019; README specifies TensorFlow 1.2.1, observed 2026-10-06.
- Behavior under transformations
- Camera processing, recompression and resizing can alter residual patterns.
- False positives and false negatives
- Benign local processing or unusual capture pipelines can appear anomalous; weak maps can be inconclusive.
- Privacy and data rights
- Local analysis preferred; learned traces can support camera-family linkage.
- Potential LegitPix role
- Research-only forensic layer pending rights and validation.
- Uncertainty and open checks
- Do not label this PRNU device identification; maintenance and production compatibility unverified.
Primary sources
passive forensicsfd-prnu
Sensor-pattern / PRNU source-camera comparison
Estimate photo-response non-uniformity from reference captures and test correlation against a questioned image or region.
What it supports
Conditional support for a shared sensor source or locally inconsistent sensor trace.
What it does not prove
Scene truth, original capture time, photographer, or universal unique-device attribution.
Implementation, limitations and sources
- Implementation and access
- Polimi ISPL prnu-python, referencing Binghamton's MATLAB implementation.
- Offline feasibility
- Yes; reference captures and careful alignment are normally required.
- Cost and dependencies
- Reference acquisition, compute, expertise and device-specific validation.
- Licenses and rights
- prnu-python MIT; other implementations/data separately licensed.
- Version or date
- Official repository observed 2026-10-06; computational-photography caution paper 2020.
- Behavior under transformations
- Resizing, stabilization, denoising, compression and computational photography affect extractability/alignment.
- False positives and false negatives
- Modern processing can introduce non-unique artifacts and false matches. Non-detection is not proof of synthetic content.
- Privacy and data rights
- Device linkage can reveal private source relationships; restrict reference databases.
- Potential LegitPix role
- Specialist source-consistency evidence with controls, not default authenticity scoring.
- Uncertainty and open checks
- Unique-sensor conclusions cannot be transferred uncritically across modern devices and modes.
Primary sources
passive forensicsfd-video-structure
Video timeline, frame correlation and container/codec forensics
Inspect timing, frame relationships, container structure, GOP/compression evidence and source-device exemplars; assess audio separately.
What it supports
Consistency or inconsistency with the claimed capture/processing history.
What it does not prove
Truth of the filmed event, lack of staging, or authenticity of an unexamined audio stream.
Implementation, limitations and sources
- Implementation and access
- SWGDE workflow plus validated file/frame inspection tools; exemplar-device comparisons where available.
- Offline feasibility
- Yes for file examination.
- Cost and dependencies
- Analyst expertise, original-quality media and source exemplars.
- Licenses and rights
- Guidance is not software; SWGDE redistribution conditions apply to its documents.
- Version or date
- SWGDE 23-V-001-1.2, 2024-03-07.
- Behavior under transformations
- Transcoding, frame-rate conversion, stabilization and legitimate editing can explain apparent discontinuities.
- False positives and false negatives
- Missing/duplicated frames need contextual explanation; heavily processed video can be inconclusive. Avoid absolute certainty.
- Privacy and data rights
- Preserve original and audit trail; restrict video/audio evidence access.
- Potential LegitPix role
- Specialist multi-signal report; never substitute temporal smoothness for authenticity.
- Uncertainty and open checks
- SWGDE lists method families; tool-specific accuracy still requires validation.
Primary sources
perceptual matchingfd-classic-perceptual-hashes
Classical perceptual image hashes: aHash, dHash, pHash, wHash, color and crop-resistant hashes
Compact descriptors summarize luminance averages, gradients, DCT/wavelet structure, color, or segmented image regions; compare using algorithm-appropriate distances.
What it supports
Candidate near-duplicate or region relationships to an indexed reference.
What it does not prove
Cryptographic integrity, common author, originality, or image truth.
Implementation, limitations and sources
- Implementation and access
- JohannesBuchner ImageHash, Pillow/NumPy/SciPy/PyWavelets dependencies as applicable.
- Offline feasibility
- Yes; CPU-friendly.
- Cost and dependencies
- Local compute and reference index.
- Licenses and rights
- ImageHash BSD-2-Clause; dependencies separately licensed.
- Version or date
- ImageHash 4.3.2, released 2025-02-01.
- Behavior under transformations
- Some tolerate moderate resizing/recompression; rotation, crops, overlays and color changes affect variants differently.
- False positives and false negatives
- Small hashes collide on simple/repetitive images. Hamming thresholds are dataset-specific; color similarity is especially weak identity evidence.
- Privacy and data rights
- A hash can reveal database membership; index only media you may retain.
- Potential LegitPix role
- Low-cost candidate retrieval followed by stronger comparison.
- Uncertainty and open checks
- Variants are not interchangeable; no universal threshold or verified robustness ranking.
Primary sources
perceptual matchingfd-local-features
Local keypoints plus geometric verification (SIFT/ORB)
Match local descriptors, then test geometrically consistent correspondences using robust estimation such as RANSAC.
What it supports
A shared visual region and an estimated geometric relationship.
What it does not prove
Whole-image identity, copying direction, authorship, or scene authenticity.
Implementation, limitations and sources
- Implementation and access
- OpenCV feature2d and homography tools; SIFT or ORB with appropriate descriptor distance.
- Offline feasibility
- Yes; CPU implementations available.
- Cost and dependencies
- Feature extraction and candidate-pair verification.
- Licenses and rights
- OpenCV 4.5+ Apache-2.0; inspect bundled dependencies.
- Version or date
- OpenCV 4.13.0 documentation.
- Behavior under transformations
- Can accommodate scale, rotation, crop and perspective within model assumptions.
- False positives and false negatives
- Repeated textures can yield false correspondences; low-texture or highly changed regions can fail. Planar homography is not universal 3D geometry.
- Privacy and data rights
- Avoid repurposing descriptors for identity tracking without an appropriate basis.
- Potential LegitPix role
- Confirm candidate region reuse with match visualizations and inlier counts.
- Uncertainty and open checks
- Tutorial thresholds are examples, not validated operating points.
Primary sources
perceptual matchingfd-pdq
PDQ image similarity hashing
Meta's PDQ produces a 256-bit perceptual signature for photo matching.
What it supports
Near-duplicate evidence relative to a particular reference and matching policy.
What it does not prove
Byte identity, safety category, copyright, or the truth of the reference.
Implementation, limitations and sources
- Implementation and access
- ThreatExchange PDQ C++ and integrations; use its quality and matching conventions.
- Offline feasibility
- Yes; hashing does not require the ThreatExchange service.
- Cost and dependencies
- Local indexing; shared reference sets are a separate access dependency.
- Licenses and rights
- ThreatExchange BSD license; repository notes file-level exceptions such as CImg.
- Version or date
- Official main branch observed 2026-10-06; commit not pinned.
- Behavior under transformations
- Designed for perceptual copies; geometric edits and major content changes remain important tests.
- False positives and false negatives
- Threshold/quality handling and corpus scale affect error rates. A reference label is not inherent in the hash.
- Privacy and data rights
- Fingerprint exchange can expose reference membership; sensitive hash sets need access controls.
- Potential LegitPix role
- Practical optional near-duplicate matcher; preserve reference ID, score and threshold.
- Uncertainty and open checks
- No independent LegitPix benchmark or installed build verified.
Primary sources
perceptual matchingfd-sscd
SSCD learned image-copy descriptors
Self-supervised image-copy model emits normalized embeddings for cosine/distance-based retrieval.
What it supports
Copy-like similarity to an indexed image, including some substantial visual alterations.
What it does not prove
Cryptographic equality, provenance chain, copyright, or semantic truth.
Implementation, limitations and sources
- Implementation and access
- Meta SSCD TorchScript/PyTorch models; typically 512-dimensional descriptors, larger variant 1024.
- Offline feasibility
- Yes after weights and dependencies are acquired; CPU or GPU.
- Cost and dependencies
- Model inference and vector index; target-domain normalization may be useful.
- Licenses and rights
- Code MIT; verify downloaded weights and all dependencies separately.
- Version or date
- CVPR 2022; main repository observed 2026-10-06.
- Behavior under transformations
- Trained for image-copy transformations; test crops, overlays, recompression and screenshots on target data.
- False positives and false negatives
- Repository's example precision/threshold is specific to DISC and its model; it is not a transferable probability.
- Privacy and data rights
- Embeddings retain linkability; external vector services transmit derived data.
- Potential LegitPix role
- Higher-capacity candidate matcher, subject to local operating-point validation.
- Uncertainty and open checks
- Research environment uses older dependencies; no maintained SLA or commercial weight clearance inferred.
Primary sources
provenanceprov-c2pa-hard
C2PA Content Credentials: signed manifests and hard bindings
Cryptographically sign claims/assertions and bind them to specified asset bytes or structured portions; actions and ingredients describe claimed history.
What it supports
Signature/claim integrity and covered-content consistency under an identified validation and trust policy.
What it does not prove
Scene truth, complete history, legal rights, or accuracy of every metadata field. Hard binding need not cover every byte.
Implementation, limitations and sources
- Implementation and access
- Open standard with multiple producer/validator implementations; manifests embedded or external. Specification 2.5 expands format and live-media coverage, adds linked-resource integrity, permits ML-DSA signatures, and clarifies binding and validation. These are published specification capabilities, not evidence that every implementation supports them.
- Offline feasibility
- Conditional: local validation with asset/manifests/trust material; fresh revocation and remote ingredients may need network.
- Cost and dependencies
- Verification can be local/free; trusted production signing, timestamps and conformance may cost.
- Licenses and rights
- Open royalty-free standard; specification content CC BY 4.0; software licenses separate.
- Version or date
- C2PA 2.5 official specification, October 2026; checked 26 October. Prior 2.4 references retained as historical sources.
- Behavior under transformations
- Changed covered content fails old hard binding; a compliant editor may issue a new signed derivative referencing ingredients.
- False positives and false negatives
- Missing, invalid, cryptographically valid and trusted are distinct states; signed assertions can be false.
- Privacy and data rights
- Manifests may expose creator, location or edit history; minimize before publication.
- Potential LegitPix role
- Primary interoperable provenance layer alongside whole-file receipt hashing; retain excluded-range semantics.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed. Exact publication day and implementation-level 2.5 support were not established; check each feature against the chosen validator release.
Primary sources
provenanceprov-jpeg-trust
JPEG Trust framework
Framework for annotating and evaluating media provenance, integrity, attribution and trust under defined policies.
What it supports
Evidence can be evaluated against a stated trust profile and reported consistently.
What it does not prove
Universal truth, complete history or a legal ownership finding.
Implementation, limitations and sources
- Implementation and access
- ISO/IEC 21617 framework; independently verify implementations and profile support.
- Offline feasibility
- Depends on validator, locally available assertions and external trust dependencies.
- Cost and dependencies
- Standards access and implementation costs require checking; none purchased.
- Licenses and rights
- ISO standard copyright and implementation rights are separate; no reference-code license verified.
- Version or date
- ISO/IEC 21617-1:2026, edition 2, published August 2026. Part 2 remains FDIS at stage 50.00; Part 3 DIS voting closed on 7 October 2026 (40.60), without publication. Part 4 remains DIS: ISO stage 40.60, voting closed 18 September 2026. This corrects the previously recorded 40.20 stage; it is not a newly published standard.
- Behavior under transformations
- Media changes require renewed integrity/provenance evaluation; not itself a robust watermark.
- False positives and false negatives
- Trust results depend on the selected profile and quality of evidence.
- Privacy and data rights
- Annotations can expose attribution or contextual data.
- Potential LegitPix role
- Track standards alignment and explicit trust-policy reporting.
- Uncertainty and open checks
- Historical observations (not current contradictions): ISO lists edition 2 as published while the JPEG workplan still describes it as in development; use ISO for publication status. Full specification and implementation conformance not evaluated. On 19 October the ISO Part 4 ?browse=tc view showed dated stage 40.60, while the canonical URL and JPEG workplan returned older 40.20 snapshots. The newer dated ISO lifecycle record is used; cache disagreement is retained explicitly. On 26 October the JPEG workplan agrees with published Part 1 edition 2 and Part 4 DIS 40.60, resolving the earlier workplan disagreement; historical observations above are retained. Rechecked 2 November: the JPEG workplan still agrees on Part 1 edition 2 published, Part 2 FDIS 50.00, and Parts 3 and 4 DIS 40.60. Full specification and implementation conformance remain unevaluated.
Primary sources
provenance / identityprov-cawg-identity
CAWG Identity Assertion
Separate signed identity assertion binds a credential holder and stated role to selected C2PA assertions, independent of the tool’s claim-signing identity.
What it supports
Control/verification of a specified digital identity and its signed relationship to selected assertions.
What it does not prove
Identity honesty, copyright ownership, complete authorship, or truth of depicted content.
Implementation, limitations and sources
- Implementation and access
- X.509/COSE and identity-claims aggregation routes; compatible validators required.
- Offline feasibility
- Conditional on packaged credentials, trust roots and status; DID/status resolution can need network.
- Cost and dependencies
- Identity provider/CA integration may require accounts, verification and fees.
- Licenses and rights
- W3C Patent Policy; specification reference code Apache 2.0 unless otherwise stated.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Follows bound assertion integrity; no extra pixel robustness.
- False positives and false negatives
- Display-name metadata and claim-generator certificate are not equivalent to verified creator identity.
- Privacy and data rights
- Opt-in identity can expose legal names, affiliations and account links permanently; minimize disclosures.
- Potential LegitPix role
- Optional attributable evidence, not a requirement for ordinary users or anonymous documentation.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
provenance / registryprov-c2pa-soft
C2PA durable credentials: soft bindings and manifest recovery
Use watermark identifiers or content fingerprints to discover a stored manifest when embedded metadata is missing.
What it supports
Probabilistic association to candidate provenance records, followed by separate signature validation.
What it does not prove
Recovered association is not a successful hard binding to a transformed file or proof that all edits were logged.
Implementation, limitations and sources
- Implementation and access
- C2PA Soft Binding API and algorithm registry; fingerprint/watermark implementations and repositories vary.
- Offline feasibility
- Only with local detector/index and cached manifests; public repository discovery normally online.
- Cost and dependencies
- Registry storage, lookup, watermark licenses and uptime are separate dependencies.
- Licenses and rights
- Open specification; algorithm/model/provider terms separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Designed to reconnect transformed assets; success depends on each watermark/fingerprint and registered coverage.
- False positives and false negatives
- False associations and absent records are possible; do not upgrade a soft match to byte-identical verification.
- Privacy and data rights
- Queries can reveal which content is being inspected; prefer hash/identifier-only or local lookup where supported.
- Potential LegitPix role
- Optional candidate-discovery stage, with visibly weaker evidence state than exact-byte integrity.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
provenance / softwareprov-cai-sdk
CAI open-source validators and signers (c2pa-rs/c2patool/JS)
Implementation layer for reading, validating and authoring C2PA evidence; different language/runtime interfaces share core functionality.
What it supports
Results of the configured C2PA validator on supported formats, not a new kind of truth claim.
What it does not prove
The implementation is bug-free or conformant merely because it uses this SDK; no scene-truth verdict.
Implementation, limitations and sources
- Implementation and access
- Rust SDK, command-line c2patool and JavaScript/WASM ecosystem; public local validation possible. c2patool v0.28.2 updates its SDK dependency to 0.91.2 and fixes external-manifest handling during export and inspection. Rust SDK 0.91.3 release notes report SVG root namespace handling and Ed25519 certificate-parameter validation fixes; C FFI 0.91.3 adds a Builder label-setting API. These changes were not independently tested. Subsequent maintainer notes: 0.91.4 changes handling of SHA-1 timestamp imprints; 0.91.5 fixes rejection of valid MOV inputs; 0.91.6 updates the ECDSA stack and c2pa_raw_crypto to 0.2.0. c2patool 0.28.3 adopts SDK 0.91.6. No fixes were independently reproduced. Maintainer release 0.91.7 updates TLS dependency minimums and requires eMudhra distrust updates; c2patool 0.28.4 adopts SDK 0.91.7. This is dependency maintenance, not a new media-authenticity guarantee.
- Offline feasibility
- Yes for local evidence; explicitly configure remote fetching, trust roots and status freshness.
- Cost and dependencies
- No SDK license fee; compute, storage, certificates and integration remain.
- Licenses and rights
- Rust/c2patool MIT and Apache-2.0; JS project MIT; transitive licenses vary.
- Version or date
- c2patool v0.28.4, c2pa Rust SDK 0.91.7 and C FFI 0.91.7 were released 26 October 2026 after the prior maintenance review. Rust GitHub tag is Pre-release; C FFI and c2patool are separately labeled Latest. The docs.rs latest page returned a cached 0.91.6 snapshot on 2 November; crate 0.91.7 publication was not independently corroborated there. Release labels do not establish C2PA 2.5 feature completeness.
- Behavior under transformations
- Format/feature support varies by release; unsupported evidence must be separate from invalid.
- False positives and false negatives
- Pin release and configuration; preserve detailed validation codes rather than a single boolean.
- Privacy and data rights
- Local-only operation requires checking remote URL and telemetry behavior, not assuming it from “open source.”
- Potential LegitPix role
- Preferred baseline for independent signature validation; no SDK or model was installed during this research.
- Uncertainty and open checks
- Release notes reviewed, not installed or independently reproduced. GitHub labels for the Rust and C FFI tags differ; a Latest tag for one component must not be generalized to every package. Pin the exact component/version and verify its supported specification features. The C FFI 0.91.7 release body repeats October 16 notes; its October 26 release date is taken from GitHub release metadata, not from that older body. The direct docs.rs 0.91.7 page could not be fetched.
Primary sources
- c2pa-rs
- c2patool
- c2pa-js documentation
- c2patool v0.28.2 release notes
- c2pa-v0.91.2 release notes
- c2pa C FFI v0.91.2 release page
- c2pa-v0.91.3 release notes
- c2pa C FFI v0.91.3 release page
- Published c2pa crate
- c2pa-v0.91.4 release notes
- c2pa-v0.91.5 release notes
- c2pa-v0.91.6 release notes
- c2pa component release listing
- c2patool v0.28.3 release notes
- c2pa-v0.91.7 release notes
- c2pa C FFI v0.91.7 release page
- c2patool v0.28.4 release notes
provenance / trustprov-c2pa-trust
C2PA trust lists, certificates and conformance
Validate certificate paths, timestamps and status against chosen trust roots; conformance evaluates a product implementation.
What it supports
A claim’s signer chains to the selected trust policy and applicable certificate conditions.
What it does not prove
A trusted signer is honest about every assertion, and conformance does not certify the depicted event.
Implementation, limitations and sources
- Implementation and access
- Official Conforming Products List, C2PA Trust List and TSA Trust List; legacy ITL remains distinguishable.
- Offline feasibility
- Cached roots/status enable bounded offline checks; report freshness and unavailable status explicitly.
- Cost and dependencies
- Production signing may require evaluation, onboarding and certificate fees; reading standards is free.
- Licenses and rights
- Public specifications/lists; onboarding and certificate contracts apply.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Independent of perceptual transformation robustness; validates cryptographic and governance evidence.
- False positives and false negatives
- Unknown trust root is not the same as invalid signature. Legacy ITL acceptance depends on validator and validity period.
- Privacy and data rights
- Certificate and status lookups can expose activity; restrict unsolicited external fetching.
- Potential LegitPix role
- Version trust policy and list snapshot in every verification receipt.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
reference matching servicefd-acoustid
AcoustID reference-database lookup
Submit a Chromaprint fingerprint and duration to retrieve matching recording IDs and optional MusicBrainz metadata.
What it supports
A database association with a known recording.
What it does not prove
Complete audio provenance, correct rights ownership, or that absent matches are original.
Implementation, limitations and sources
- Implementation and access
- Documented v2 lookup API; application key required.
- Offline feasibility
- Hosted lookup no; self-hosted software/data is a separate engineering path.
- Cost and dependencies
- Public endpoint is noncommercial with 3 requests/second guideline; commercial use requires the commercial service.
- Licenses and rights
- Service software MIT; fingerprint/metadata database CC BY-SA 3.0; API terms separate.
- Version or date
- Official service, licensing and commercial pages observed 2026-10-06.
- Behavior under transformations
- Inherits fingerprint and catalog-coverage limitations.
- False positives and false negatives
- Crowdsourced metadata can be wrong or ambiguous; service score is not an authenticity probability.
- Privacy and data rights
- Fingerprint/duration queries leave the device; submission is a separate data-sharing action.
- Potential LegitPix role
- Optional external reference lookup with explicit user data-sharing authorization.
- Uncertainty and open checks
- No query, subscription, account creation or fingerprint submission performed.
Primary sources
reference matching servicefd-acrcloud
Commercial audio/reference recognition (ACRCloud representative)
Commercial API accepts audio or SDK-generated fingerprints and returns catalog/custom-reference recognition metadata.
What it supports
A provider-scoped reference match; some products return segment offsets.
What it does not prove
Authenticity, voice identity, ownership, or image/video-frame verification merely because the container is video.
Implementation, limitations and sources
- Implementation and access
- ACRCloud identification endpoint with authenticated project credentials; custom buckets and product-specific metadata.
- Offline feasibility
- Hosted identification requires network; docs mention offline project types, but licensing/deployment was not verified.
- Cost and dependencies
- Account, commercial plan, SDK and selected catalog/product.
- Licenses and rights
- Proprietary service/SDK terms; public client examples do not license the matching engine.
- Version or date
- Official API documentation observed 2026-10-06.
- Behavior under transformations
- Product-specific matching robustness must be tested; avoid extrapolating provider marketing.
- False positives and false negatives
- Catalog coverage, content duration and shared audio affect interpretation; provider scores need calibration.
- Privacy and data rights
- Uploading audio or fingerprints transmits potentially private recording information.
- Potential LegitPix role
- Optional adapter only after rights, disclosure, cost and data-processing review.
- Uncertainty and open checks
- No account, pricing quote, offline entitlement or accuracy validation obtained.
Primary sources
reference matching servicefd-private-matching
Hosted private image-reference matching (TinEye MatchEngine representative)
Upload an authorized image collection, create provider fingerprints, and query that private index.
What it supports
Similarity to an item in the supplied reference collection.
What it does not prove
Public-web discovery, creator identity, ownership or scene truth.
Implementation, limitations and sources
- Implementation and access
- MatchEngine API; catalog import and query workflow.
- Offline feasibility
- No public offline deployment established.
- Cost and dependencies
- Paid service, storage/index scope and vendor agreement.
- Licenses and rights
- Proprietary API/service terms.
- Version or date
- Official MatchEngine page observed 2026-10-06.
- Behavior under transformations
- Provider advertises matching resized, cropped, color-changed and occluded variants; noisy mobile photos are a separate product focus.
- False positives and false negatives
- Vendor robustness statements are not independent validation; threshold and reference-set composition matter.
- Privacy and data rights
- Provider explicitly stores submitted reference images on its servers; approval and retention/access controls are required.
- Potential LegitPix role
- Commercial alternative to self-hosted matching when data sharing is acceptable.
- Uncertainty and open checks
- No quote, deployment test or contractual privacy/security assessment performed.
Primary sources
reference matching servicefd-photodna
PhotoDNA known-image matching
Microsoft's perceptual signatures match copies against controlled reference sets of previously identified material.
What it supports
A candidate known-reference match within the approved program.
What it does not prove
General image authenticity, identity of a depicted person, or new/unknown content classification.
Implementation, limitations and sources
- Implementation and access
- PhotoDNA Cloud Service and approved tool-provider integrations.
- Offline feasibility
- Not a publicly available general-purpose offline SDK; qualified access/integration must be established.
- Cost and dependencies
- Microsoft says free for qualified customers, law enforcement and tool providers; eligibility applies.
- Licenses and rights
- Proprietary program/service terms.
- Version or date
- Official overview observed 2026-10-06.
- Behavior under transformations
- Designed to find copies; precise current performance envelope not independently verified.
- False positives and false negatives
- Reference coverage and policy determine meaning. No match cannot establish safety or authenticity.
- Privacy and data rights
- Highly sensitive safety workflows require specialized access, governance and lawful handling; do not build a private illegal-media reference set.
- Potential LegitPix role
- Taxonomy/access-boundary reference only, not a default LegitPix integration.
- Uncertainty and open checks
- No eligibility, SDK access or service evaluation verified.
Primary sources
reference matching servicefd-content-id
YouTube Content ID audiovisual reference matching
YouTube matches uploads to audio/visual reference files submitted by eligible rights holders.
What it supports
Platform-reported reuse of reference content.
What it does not prove
Authenticity, legal infringement, valid ownership, or comprehensive internet coverage.
Implementation, limitations and sources
- Implementation and access
- YouTube partner rights-management workflow, not a public authenticity API.
- Offline feasibility
- No public offline implementation.
- Cost and dependencies
- Eligibility, exclusive rights, substantial catalog/need, geographic rights information and agreement.
- Licenses and rights
- Proprietary access terms; reference rights restrictions.
- Version or date
- Official help pages observed 2026-10-06.
- Behavior under transformations
- Audio and visual matching behavior is proprietary; do not assume all edited reuse is detected.
- False positives and false negatives
- Erroneous claims occur and disputes exist. A Content ID claim is a rights-management event, not a legal adjudication.
- Privacy and data rights
- Uploading media and enabling claims can have consequential publishing/rights effects.
- Potential LegitPix role
- Documented external evidence if legitimately available; never use upload as an unapproved analysis workaround.
- Uncertainty and open checks
- No access or partnership assumed; no test upload made.
Primary sources
registry / signatureregistry-sigstore
Sigstore/Cosign/Rekor transparency evidence
Sign an artifact or receipt; bind short-lived certificates to an authenticated identity and log signing evidence in an append-only transparency service.
What it supports
Expected signer identity signed the committed artifact and evidence is included in the selected log.
What it does not prove
Signer assertions are true, a camera captured the file, copyright ownership, or a globally complete history.
Implementation, limitations and sources
- Implementation and access
- Public Cosign blob signing/verification, Fulcio CA and Rekor; bundle-based evidence verification.
- Offline feasibility
- Possible with complete bundle, trust root and required proof material; issuance/log submission online.
- Cost and dependencies
- Public-good services or self-hosted infrastructure; identity-provider availability and operational policies matter.
- Licenses and rights
- Cosign Apache-2.0; service policies and other component licenses separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Exact artifact/receipt signature; edits require a new signed object.
- False positives and false negatives
- A signed inclusion promise differs from an independently verified inclusion/consistency proof; verify expected identity and issuer.
- Privacy and data rights
- Public logs can permanently expose identity/email and digest. Never publish without explicit consent.
- Potential LegitPix role
- Potential future auditable receipt infrastructure; not an automatic replacement for privacy-preserving local receipts.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
reverse searchfd-google-lens
Public reverse/visual image search (Google Lens representative)
Submit an image or image URL to discover visually related images and websites.
What it supports
A discovery lead linking visible content to indexed pages.
What it does not prove
Original source, earliest publication, truthful caption, license, or absence elsewhere.
Implementation, limitations and sources
- Implementation and access
- Google Lens browser/search interfaces; no general offline equivalent.
- Offline feasibility
- No.
- Cost and dependencies
- Hosted index, availability and interface limitations.
- Licenses and rights
- Proprietary service terms; results retain third-party rights.
- Version or date
- Official help page observed 2026-10-06.
- Behavior under transformations
- Can return similar subjects as well as pages with the same image.
- False positives and false negatives
- Visual matches and AI summaries are leads, not verification. Index omissions, removed pages and changed context matter.
- Privacy and data rights
- Image/URL is shared with Google; its help warns searched URLs may be stored.
- Potential LegitPix role
- Opt-in source discovery followed by opening and corroborating the underlying page.
- Uncertainty and open checks
- No media upload or search benchmark performed; coverage is opaque.
Primary sources
reverse searchfd-tineye
Reverse-copy web search (TinEye)
Fingerprint an image and search a crawled web index for copies and edited variants.
What it supports
An indexed occurrence or candidate modified copy.
What it does not prove
Creator, ownership, first publication, authenticity or originality after no match.
Implementation, limitations and sources
- Implementation and access
- TinEye web search; separate paid API product.
- Offline feasibility
- No for the public web index.
- Cost and dependencies
- Service availability, indexed coverage and API contract for automation.
- Licenses and rights
- Proprietary service/API terms; source-page content separately licensed.
- Version or date
- Official explanations updated 2022-11-08 and 2025-02-24.
- Behavior under transformations
- Finds cropped, edited, resized and partial copies; not usually different pictures of the same subject.
- False positives and false negatives
- Indexed/crawl dates are not capture dates. Verify page context and compare the actual images.
- Privacy and data rights
- Provider says search images are briefly retained and not added to its index; upload still requires authorization.
- Potential LegitPix role
- Optional copy-discovery evidence with returned URL and observation time.
- Uncertainty and open checks
- No service query performed; retention claims are provider statements.
Primary sources
semantic retrievalfd-semantic-embeddings
Semantic image/text embeddings (CLIP-style retrieval)
Encode images and descriptions into a shared representation to retrieve semantically related content.
What it supports
Similarity in model representation; useful leads for scene or caption investigation.
What it does not prove
Same underlying asset, accurate caption, identity, event, place or capture time.
Implementation, limitations and sources
- Implementation and access
- OpenAI CLIP or separately reviewed compatible implementations.
- Offline feasibility
- Yes after weights are obtained.
- Cost and dependencies
- Model inference and vector search.
- Licenses and rights
- Official CLIP code/model release uses MIT; third-party checkpoints may differ.
- Version or date
- CLIP model card January 2021, variants through April 2022.
- Behavior under transformations
- Semantic similarity may persist despite major visual differences, which is not evidence of copying.
- False positives and false negatives
- Shared subject matter can produce high similarity for unrelated images. Model bias and out-of-domain behavior require evaluation.
- Privacy and data rights
- Do not use retrieval embeddings as an unreviewed biometric or sensitive-trait classifier.
- Potential LegitPix role
- Discovery-only signal; never substitute its similarity for a copy hash.
- Uncertainty and open checks
- Original model card says deployment needs careful context-specific study.
Primary sources
timestamptime-rfc3161
RFC 3161 trusted timestamps
A time-stamping authority signs a digest and time under a stated certificate/policy; RFC 5816 updates the protocol.
What it supports
The committed datum existed no later than the trusted timestamp, conditional on TSA/hash security.
What it does not prove
Original capture/creation time, authorship, truth, ownership, or public publication time.
Implementation, limitations and sources
- Implementation and access
- Standard TSA clients/servers; evidence can accompany a signature or receipt.
- Offline feasibility
- Verify stored token with trusted certificates/status evidence; issuance requires TSA or internal trusted service.
- Cost and dependencies
- TSA pricing varies; self-operated TSA creates its own trust burden.
- Licenses and rights
- Open protocol; software licenses and TSA contracts separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Exact digest commitment; changed bytes need a new timestamp or authenticated derivation.
- False positives and false negatives
- Local computer time is not a trusted timestamp; check token policy, imprint, certificate and validity.
- Privacy and data rights
- Only digest is needed; digest disclosure can still link a known file to a timestamp request.
- Potential LegitPix role
- Useful optional time anchor for local receipts; state “existed by” rather than “created at.”
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
timestamp / registrytime-opentimestamps
OpenTimestamps Bitcoin anchoring
Aggregate hash commitments into Bitcoin and retain a proof linking a file digest to a block.
What it supports
Committed bytes existed before a blockchain-anchored time bound under hash/chain assumptions.
What it does not prove
Who created the file, exact capture time, ownership, truthful content or that an incomplete proof is anchored.
Implementation, limitations and sources
- Implementation and access
- Public browser/Python clients and calendar servers; .ots evidence file.
- Offline feasibility
- Complete proofs can verify locally with required Bitcoin chain data; Python client documents local Bitcoin Core.
- Cost and dependencies
- Public calendars advertise free use without registration; direct anchoring has network fees. Confirmation may take hours.
- Licenses and rights
- Python client LGPLv3-or-later; other implementations have separate terms.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Exact hash binding; any changed file fails unless independently committed.
- False positives and false negatives
- Pending calendar attestation is not completed Bitcoin proof; archive upgraded proof and verifier context.
- Privacy and data rights
- Local hashing avoids raw upload; commitments and network metadata still have correlation risk.
- Potential LegitPix role
- Optional independent existence anchor, not required for basic local byte receipts.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
video matchingfd-tmk
TMK+PDQF temporal video matching
Temporal Match Kernel combines visual frame features into a fixed-size video signature.
What it supports
Whole-video similarity to a known reference under the implementation's alignment/scoring model.
What it does not prove
Framewise integrity, exact clip inclusion, or authenticity of the depicted event.
Implementation, limitations and sources
- Implementation and access
- ThreatExchange TMK+PDQF; decoding dependencies and optional vector indexing.
- Offline feasibility
- Yes.
- Cost and dependencies
- Video decoding, signature storage and search.
- Licenses and rights
- ThreatExchange BSD with documented file/dependency exceptions.
- Version or date
- Official repository observed 2026-10-06.
- Behavior under transformations
- Optimized for videos of comparable/same length; frame-feature limitations also apply.
- False positives and false negatives
- Official comparison says TMK is not the subsequence/clip-matching alternative. Sampling and temporal edits can change results.
- Privacy and data rights
- Store authorized reference signatures with source provenance.
- Potential LegitPix role
- Optional whole-video relationship evidence, distinctly labeled from vPDQ clip matches.
- Uncertainty and open checks
- No benchmark/runtime verification performed for LegitPix.
Primary sources
video matchingfd-vpdq
vPDQ frame-overlap and clip matching
Hash sampled frames using PDQ and compare shared similar frames, permitting subsequence/segment matches.
What it supports
A candidate shared visual segment or frame relationship.
What it does not prove
Identical ordering/content of all frames, audio integrity, or event authenticity.
Implementation, limitations and sources
- Implementation and access
- ThreatExchange vPDQ C++/Python, video libraries; FAISS integration documented.
- Offline feasibility
- Yes after build; official README says wheels are not distributed.
- Cost and dependencies
- Variable-length fingerprints; decoding, indexing and candidate verification costs.
- Licenses and rights
- ThreatExchange BSD with dependency/file-level exceptions.
- Version or date
- Official main README observed 2026-10-06.
- Behavior under transformations
- Supports clip/subsequence reuse; inherits PDQ's image-level strengths and weaknesses.
- False positives and false negatives
- Shared title cards, static frames or repeated scenes are weak evidence alone; report coverage and time spans.
- Privacy and data rights
- Frame-level reference matching remains linkable derived data.
- Potential LegitPix role
- Segment relationship evidence plus frame matches, not a binary authenticity verdict.
- Uncertainty and open checks
- Coverage, sampling rate and thresholds must be recorded and validated.
Primary sources
watermarkwm-atsc-a334
ATSC A/334 audio watermark emission
Standardized VP1 watermark embeds packets in PCM audio for broadcast and redistribution workflows.
What it supports
Recovery of an embedded broadcast payload by a compatible extractor.
What it does not prove
AI origin, scene truth, trustworthy authorship or an unmodified recording.
Implementation, limitations and sources
- Implementation and access
- ATSC 3.0 audio watermark format; specialist embedder/extractor required.
- Offline feasibility
- Signal extraction can be local in an implementation; content recovery may require a service.
- Cost and dependencies
- Implementation and service pricing not verified.
- Licenses and rights
- Publicly downloadable specification; potential patent licenses explicitly noted. Not an open-source software grant.
- Version or date
- A/334:2026-04, approved 14 April 2026.
- Behavior under transformations
- Designed to carry data through audio distribution, compression and channel mixing; validate the intended path.
- False positives and false negatives
- Emission is optional. Unmarked and unsupported audio are not evidence of synthesis.
- Privacy and data rights
- Associated identifiers or viewing measurements require appropriate data controls.
- Potential LegitPix role
- Specialist broadcast recognition option, outside the initial photo-receipt core.
- Uncertainty and open checks
- No implementation, patent terms or detector performance independently tested.
Primary sources
watermarkwm-atsc-a335
ATSC A/335 video watermark emission
Broadcast video watermark carries ancillary data through image pixels for receiver content recovery.
What it supports
Extraction of an associated broadcast payload from compatible marked video.
What it does not prove
Indelible marking, tamper resistance, authentic footage or an ownership determination.
Implementation, limitations and sources
- Implementation and access
- ATSC 3.0 format, used with A/336 content-recovery signaling.
- Offline feasibility
- An extractor can process video locally; resolving payloads may need broadband services.
- Cost and dependencies
- Implementation, patent and service costs unverified.
- Licenses and rights
- Downloadable specification with patent-rights warning; implementation rights need separate review.
- Version or date
- A/335:2026-04, approved 14 April 2026.
- Behavior under transformations
- Designed for compression-rate changes, codec transcoding and HDMI redistribution; explicitly not tamper-resistant.
- False positives and false negatives
- Optional emission and unsupported signal conditions make absence inconclusive.
- Privacy and data rights
- Content IDs can support viewing linkage; review receiver telemetry separately.
- Potential LegitPix role
- Document the distinction between content-recognition signaling and authenticity evidence.
- Uncertainty and open checks
- No implementation or detection benchmark performed.
Primary sources
watermarkwm-audio-echo
Audio echo hiding
Encode bits in low-level echo characteristics. Perceptual masking helps hide added signal, with recovery based on the echo structure.
What it supports
Recovery of an expected audio payload under the configured channel and decoder.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Classical published signal-processing family; no maintained original production package verified.
- Offline feasibility
- In principle yes, with conventional CPU signal processing.
- Cost and dependencies
- Implementation/integration work; no neural-weight service required.
- Licenses and rights
- Publication does not establish an implementation or patent license; no legal status assessment.
- Version or date
- 1996 foundational research; current-source review October 6, 2026
- Behavior under transformations
- Channel reverberation, filtering and duration influence recoverability; no universal robustness number established.
- False positives and false negatives
- Presence thresholds and synchronization matter. A hidden message is not authenticated merely because it decodes.
- Privacy and data rights
- Local implementation possible; payload identifiers should minimize personal data.
- Potential LegitPix role
- Historical/classical audio baseline; compare against modern local tools before considering deployment.
- Uncertainty and open checks
- Primary published mechanism verified; current production readiness, perceptual tests and error rates not independently reproduced.
Primary sources
watermarkwm-audio-phase
Audio phase coding
Encode data through controlled phase changes while preserving relative phases to limit audible distortion. Receiver synchronization and parameter agreement are essential.
What it supports
Recovery of an expected audio payload under the configured channel and decoder.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Classical published signal-processing family; no maintained original production package verified.
- Offline feasibility
- In principle yes, with conventional CPU signal processing.
- Cost and dependencies
- Implementation/integration work; no neural-weight service required.
- Licenses and rights
- Publication does not establish an implementation or patent license; no legal status assessment.
- Version or date
- 1996 foundational research; current-source review October 6, 2026
- Behavior under transformations
- Original experiments concern a specified signal/decoder setup; modern codecs, time changes and recording channels require new testing.
- False positives and false negatives
- Presence thresholds and synchronization matter. A hidden message is not authenticated merely because it decodes.
- Privacy and data rights
- Local implementation possible; payload identifiers should minimize personal data.
- Potential LegitPix role
- Historical/classical audio baseline; compare against modern local tools before considering deployment.
- Uncertainty and open checks
- Primary published mechanism verified; current production readiness, perceptual tests and error rates not independently reproduced.
Primary sources
watermarkwm-audioseal
AudioSeal
Neural waveform watermark with sample-level localization and optional 16-bit message; detection and message extraction are separate outputs.
What it supports
Detected portions carry an AudioSeal-style mark; optional payload identifies configured source/version.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Public generator/detector and training code; v0.2 adds streaming.
- Offline feasibility
- Yes after weights; CPU/GPU performance depends on length.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code and weights since April 2, 2024.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Tests cover compression, re-encoding and noise; speech/rate/channel assumptions matter.
- False positives and false negatives
- Short/mixed clips and threshold choice matter; 16-bit identifiers are not globally unique or cryptographic authentication.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Candidate local audio localization adapter; preserve per-segment scores and expected-message check.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-audiowmark
audiowmark (spectral patchwork)
Traditional spectral patchwork audio marking with error correction and optional keyed operation.
What it supports
Recovery of an expected 128-bit identifier without original audio.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- C++ CLI; companion video wrapper marks the audio track, including HLS use.
- Offline feasibility
- Yes, conventional CPU/FFT dependencies; no neural weights.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- GPLv3-or-later; assess distribution/linking obligations.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Project reports usual recovery after MP3/Ogg at 128 kbps or above in its tests; other channels require evaluation.
- False positives and false negatives
- Short duration and recording channel affect recovery; project recommends hash/HMAC-style payloads.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Non-neural local audio baseline with explicit copyleft review.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-chunkyseal
ChunkySeal
Scaled-up learned watermark emphasizing larger payload capacity.
What it supports
Recovery of up to 1024 embedded bits under tested conditions.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Model card/checkpoint in Meta VideoSeal repository.
- Offline feasibility
- Yes, after obtaining dependencies and weights; local performance untested.
- Cost and dependencies
- Larger local model; benchmark memory and throughput.
- Licenses and rights
- MIT code and models per repository.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Paper reports fourfold capacity over 256-bit baseline with measured quality/robustness; not a bound for arbitrary edits.
- False positives and false negatives
- Capacity is not authentication strength. Reserve payload for opaque identifier or authenticated data.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Capacity-oriented research candidate; signed metadata can usually carry richer context without pixel payload pressure.
- Uncertainty and open checks
- Paper demonstrates image capacity. A video-capable repository/card is not independent video-performance validation; no model run here.
Primary sources
watermarkwm-digimarc
Digimarc Validate Media and watermark-linked recovery
Commercial covert watermark identifier linked to content ownership/provenance records and detection services.
What it supports
Detected content identifier associates media with a provider-managed record.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Commercial image/audio protection and validation API; public MIT-licensed beta Chrome extension recovers supported watermark-linked C2PA via SBR endpoints.
- Offline feasibility
- Browser C2PA checks exist; watermark-linked recovery uses external Digimarc/Adobe SBR services. General offline commercial detector access not verified.
- Cost and dependencies
- Contact vendor; public unit price not verified.
- Licenses and rights
- Validate services are proprietary; separate public browser-extension code is MIT.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Vendor describes robust identifiers; detailed independent operating curves and coverage not verified here.
- False positives and false negatives
- A vendor record is an assertion, not automatic copyright ownership or factual authenticity.
- Privacy and data rights
- Browser extension processes media URLs/bytes and uses external SBR endpoints when enabled; broad site permissions and automatic scanning require careful consent.
- Potential LegitPix role
- Vendor adapter only when a real user need and access justify cost; not baseline dependency.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-fragile-spatial
Fragile spatial/LSB authentication watermarks
Embed authentication bits in pixels, often blockwise low-bit data. Fragility intentionally reveals changes; hierarchical dependencies can address weaknesses of independent blocks.
What it supports
Consistency with an authenticated marked version, sometimes localized to blocks.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Published public-key and hierarchical designs; no maintained production package verified.
- Offline feasibility
- Yes; basic image processing plus cryptographic key management.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- Paper rights retained; implementation and patent position not established.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Compression, resampling and ordinary edits may intentionally invalidate the mark; unsuitable as a durable internet identifier.
- False positives and false negatives
- Change detected does not identify malicious intent. Historic independent-block schemes have published forgery weaknesses.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Controlled lossless pipelines; distinguish tamper-localization evidence from robust attribution.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-synthid-google
Google SynthID for images, video and audio
Provider-integrated imperceptible media marks, using a learned image encoder/decoder and modality-specific systems. Visual marking can be post-hoc inside a generation workflow.
What it supports
Supported content carries a recognized SynthID mark from Google or supported partners; generated or AI-edited content may be marked.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Public SynthID Detector for supported images, video and audio. Google names OpenAI, NVIDIA and Kakao as supported partners; Apple support is forthcoming. Built-in Google product checks have their own documented coverage.
- Offline feasibility
- No public standalone media detector/weights verified.
- Cost and dependencies
- Service/account and regional availability dependencies; no local licensing or API quote established.
- Licenses and rights
- Proprietary media models; public text code does not license image/audio/video detectors.
- Version or date
- 2026-10-07 public Detector expansion; official sources reviewed 2026-10-12.
- Behavior under transformations
- Google documents common-edit resilience; technical paper evaluates external SynthID-O, not every production detector/model.
- False positives and false negatives
- No detected mark does not mean human-made; marks may indicate editing rather than wholly synthetic creation.
- Privacy and data rights
- Service checks require media upload; authorization and retention review needed. No media uploaded in this research.
- Potential LegitPix role
- Optional provider-specific external check with explicit provenance scope and privacy gate.
- Uncertainty and open checks
- Public availability is a vendor announcement, not an independent accuracy or account-access test. No media uploaded; no public standalone media weights or general third-party detector API verified. Do not treat forthcoming Apple support as released.
Primary sources
watermarkwm-imatag
IMATAG watermarking and authenticity
Commercial image/video marking; default key-based detection and optional payload modes support tracing and association.
What it supports
Match to a registered watermark/key and associated customer record.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Documented REST API v4; SaaS/API/on-premise offerings.
- Offline feasibility
- On-premise offered; exact detector, telemetry and license conditions unverified.
- Cost and dependencies
- Authenticity/Leaks pricing on request; Monitor advertised from EUR 299/month, not an authenticity API price.
- Licenses and rights
- Proprietary commercial software/service.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Robustness is configuration/service-specific; no independently verified general rate quoted.
- False positives and false negatives
- Default detection needs the proper key/registration; absent match is not evidence of non-AI origin.
- Privacy and data rights
- Vendor workflow may require registration and uploaded media; no account created or content submitted.
- Potential LegitPix role
- Optional customer-specific traceability connector, subject to licensing/privacy review.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-invisible-watermark
invisible-watermark (DWT/DCT, SVD, RivaGAN port)
Practical blind image watermark library covering classical transform variants and a learned model.
What it supports
Recovery of a caller-selected payload, subject to verification.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- ShieldMnt Python package and CLI; experimental CPU-only implementation.
- Offline feasibility
- Yes; basic methods CPU-feasible; learned/SVD routes slower.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT repository.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Repository reports JPEG/noise/brightness successes, weaker geometric resilience and poor homogeneous screenshots/posters.
- False positives and false negatives
- Authors warn even untouched media may not decode perfectly. README toy tests are not deployment-wide accuracy estimates.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Low-cost baseline only; not a generic watermark scanner or production reliability claim.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-pixelseal
PixelSeal
Adversarial-only training and high-resolution adaptation improve the quality/robustness trade-off of image/video watermarks.
What it supports
Expected 256-bit payload recovery with the released model.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Checkpoint and inference in Meta VideoSeal repository.
- Offline feasibility
- Yes, after obtaining dependencies and weights; local performance untested.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code and models per repository.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Authors evaluate many visual transformations; JND attenuation and temporal pooling address high resolution/video.
- False positives and false negatives
- “State of the art” is benchmark-relative; public detector/encoder availability is not unforgeability.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Promising permissive local baseline; benchmark on LegitPix media before adoption.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-qim
Quantization index modulation (QIM/DC-QIM)
Encode bits through quantizer choices in samples or transform coefficients. Distortion compensation adjusts the robustness/capacity trade-off.
What it supports
A payload matching an expected codebook/key was recovered.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Algorithm family implementable in image/audio/video pipelines.
- Offline feasibility
- Yes; no neural weights required.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- Primary paper does not supply a universal software/patent grant.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Provable results concern stated bounded-distortion or probabilistic channel models; real codecs and synchronization errors require measurement.
- False positives and false negatives
- Theoretical optimality does not imply robustness to arbitrary editing or forged attribution.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Optional method-specific evidence; record detector/version, key scope, threshold, payload check and limitations.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-reversible
Reversible watermarking/data hiding
Embed payload while retaining enough information to reconstruct the original pixels exactly under the method’s conditions; difference expansion is one example.
What it supports
Exact recovery of the pre-embedding image when reversibility conditions hold.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Tian difference expansion; histogram/prediction variants form a larger family.
- Offline feasibility
- Yes; CPU-friendly algorithms exist.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- Paper rights retained; no production software license verified.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Reversible does not mean durable: lossy transformations can destroy payload and exact recovery.
- False positives and false negatives
- Recoverability alone authenticates neither the payload nor the creator; add cryptographic checks.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Archival/controlled workflows where altering original pixels is unacceptable after extraction.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-rivagan
RivaGAN
Attention-based video encoder/decoder with adversarial networks balancing quality and robustness.
What it supports
A known 32-bit watermark message recovered from supported video.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Official DAI-Lab Python implementation exposes train, encode and decode.
- Offline feasibility
- Possible with trained model; old Python requirements increase maintenance cost.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code; training datasets separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Authors report robustness to common processing; no universal current-codec guarantee.
- False positives and false negatives
- Calibrate on unmarked negatives and transformed positives; decoding bits alone is not proof of a valid expected payload.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Legacy video-watermark baseline; image ports must be evaluated separately.
- Uncertainty and open checks
- Official research targets video; image-port coverage is separately catalogued under invisible-watermark. No runtime reproduction.
Primary sources
watermarkwm-semi-fragile
Semi-fragile authentication and SARI
Accept a defined class of benign processing while rejecting other changes; SARI combines localization with approximate recovery data.
What it supports
Media conforms to the scheme’s allowed-change policy.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Lin–Chang JPEG semi-fragile and SARI prototypes; current supported distribution unverified.
- Offline feasibility
- Yes, after obtaining dependencies and weights; local performance untested.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- No current software license verified.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Representative design tolerates JPEG quantization only to a predetermined quality boundary; other transformations need separate evaluation.
- False positives and false negatives
- Benign/malicious is an engineered policy, not a semantic judgment; approximate recovery is not recovered original evidence.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Specialized controlled workflows; avoid a generic “not manipulated” verdict.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-session-forensic
Session-specific forensic video watermarking (NexGuard)
Embed a deliberately assigned identifier into each distributed session/copy. This “forensic fingerprint” is not a content-derived perceptual fingerprint; tracing requires the issuance record.
What it supports
Recovered identifier corresponds to an issued copy or session.
What it does not prove
Who personally leaked the copy, scene truth, ownership, or that a subscriber intentionally acted.
Implementation, limitations and sources
- Implementation and access
- NAGRA NexGuard Streaming integrates server-side with CDNs for live and VOD.
- Offline feasibility
- Not a public standalone detector; depends on vendor/session records.
- Cost and dependencies
- Negotiated enterprise service and CDN/distribution integration.
- Licenses and rights
- Proprietary commercial service.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Vendor robustness claims require service-specific validation; collusion and capture channels belong in the threat model.
- False positives and false negatives
- Do not turn a session match into an accusation against a person without corroboration.
- Privacy and data rights
- Session identifiers can identify viewers; treat mappings as restricted personal data.
- Potential LegitPix role
- Separate leak-tracing use case, outside baseline authenticity receipt scope.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-silentcipher
SilentCipher
Neural audio watermark with psychoacoustic thresholding and compression-aware training; 40-bit messages.
What it supports
Recovery of a recognized audio payload.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Sony Python inference for 16 kHz and 44.1 kHz; downloadable models; CPU/CUDA paths.
- Offline feasibility
- Yes, after obtaining dependencies and weights; local performance untested.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code; model distribution terms should be pinned separately.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Compression and audio distortions evaluated; crop-aware decoding has additional runtime cost.
- False positives and false negatives
- Changing embedding strength can alter detection accuracy; “inaudible” depends on content/listening conditions.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- High-sample-rate audio comparison candidate, subject to model-license review.
- Uncertainty and open checks
- Repository still describes training-code release as forthcoming; no independent runtime reproduction.
Primary sources
watermarkwm-stable-signature
Stable Signature
Fine-tune a latent-diffusion decoder so generated output carries a fixed learned watermark.
What it supports
Consistency with a configured watermarked decoder/key.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Meta training, extraction and evaluation code; 48-bit extractor checkpoints.
- Offline feasibility
- Yes for licensed model owners; training/generation typically GPU-heavy.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- Predominantly CC-BY-NC; some ldm/taming code MIT; base-generator licenses separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Alternative extractors trade image quality against blur/rotation robustness; transformed outputs need evaluation.
- False positives and false negatives
- Watermarked-model attribution is conditional on secret/control assumptions; released keys/models are not exclusive author credentials.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Generator-side research comparator, not a scanner for all diffusion images.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-stegastamp
StegaStamp
Neural image watermark designed for print/photograph recovery; encodes small messages plus error correction.
What it supports
Recovery of an embedded message through the evaluated physical imaging channel.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Official TensorFlow code, weights and image/video detector; default 100 coded bits yield 56 payload bits.
- Offline feasibility
- Yes after weights; legacy TensorFlow 1.13 environment may need isolation.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT repository; check bundled dependencies.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Physical print-camera and perspective/photometric changes evaluated; detector size sensitivity and noisy region detection documented.
- False positives and false negatives
- Calibrate on unmarked negatives and transformed positives; decoding bits alone is not proof of a valid expected payload.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Optional durable locator for print/recapture scenarios, never evidence that the photographed scene itself occurred.
- Uncertainty and open checks
- Detector training code is not fully included; runtime and real-world recapture performance not reproduced.
Primary sources
watermarkwm-spread-spectrum
Transform-domain spread-spectrum watermarking
Distribute keyed low-amplitude signal across perceptually important spectral components; correlation tests retrieve presence or payload.
What it supports
A signal consistent with the expected watermark/key is present.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Cox et al. foundational DCT/spectral design; many derivatives.
- Offline feasibility
- Yes; CPU implementations feasible; original/reference requirements vary.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- Method publication is not a blanket implementation/patent license.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Paper evaluates signal processing and geometric changes; geometric robustness assumes successful registration and available original in its design.
- False positives and false negatives
- Correlation threshold, key count and original access affect false matches; do not transfer one design’s guarantees to the whole family.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Optional method-specific evidence; record detector/version, key scope, threshold, payload check and limitations.
- Uncertainty and open checks
- Source demonstrates image watermarking and proposes generalization to audio/video; those modalities need specific implementations and evidence.
Primary sources
watermarkwm-trustmark
TrustMark
Learned arbitrary-resolution image watermark with configurable error correction and public encoder/decoder.
What it supports
Expected payload recovery; can locate a separate C2PA manifest.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Adobe implementation includes Python, JavaScript and Rust routes; models downloaded on first use.
- Offline feasibility
- Yes, after obtaining dependencies and weights; local performance untested.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT explicitly covers code and model files.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Targets common visual transformations. Project also publishes a removal model, so robustness must not be read as irreversible or unremovable.
- False positives and false negatives
- A very recent single preprint reports ECC-only presence false alarms of 5.9–15.3% for evaluated variants. Verify expected payload, not merely successful ECC decode.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Strong local-integration candidate after calibrated expected-payload validation; separate soft association from hard integrity.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-videoseal
VideoSeal
Post-hoc image/video watermark using efficient temporal propagation and aggregation.
What it supports
Expected payload recovered from visual frames.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Meta PyTorch library; v1.0 256-bit stable model; streaming inference and evaluation scripts.
- Offline feasibility
- Yes; PyTorch/codec dependencies; streaming avoids loading long videos fully.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT model/code; third-party baselines have their own terms.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Evaluations cover visual alterations and video compression; visual model does not authenticate audio track.
- False positives and false negatives
- Calibrate on unmarked negatives and transformed positives; decoding bits alone is not proof of a valid expected payload.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Candidate local video adapter. Keep visual and audio evidence separate.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-visible
Visible overlays and audible disclosure marks
Human-perceptible logo, text, pattern or audible cue accompanies content. Can carry an identifier pointing to a separate receipt.
What it supports
A displayed attribution or disclosure claim exists.
What it does not prove
Who applied the label, whether its claim is correct, file integrity, scene truth or legal ownership.
Implementation, limitations and sources
- Implementation and access
- Ordinary image/video compositors and audio editors; no specialized detector.
- Offline feasibility
- Yes; low-cost conventional processing.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- Implementation-specific; no universal license.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Visibility can survive sharing but may be obscured or lost in derivatives; no cryptographic resistance.
- False positives and false negatives
- A forged or copied label looks legitimate; absence has no AI-origin meaning.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Disclosure/wayfinding only. Do not elevate an overlay to verified provenance.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-watermark-anything
Watermark Anything (WAM)
Localized neural watermarking can embed different messages in image regions and recover location/payload maps.
What it supports
A region carries a detected message; may support tracing composites.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Meta code and SA-1B checkpoint; original paper COCO checkpoint separate.
- Offline feasibility
- Yes with selected checkpoint; maintenance responsibility is local.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code/new SA-1B weights; original COCO weights CC-BY-NC 4.0.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Designed for localized crops/composites and transformed regions; mask accuracy depends on scale and calibration.
- False positives and false negatives
- Region localization is probabilistic; detector output is not an exhaustive edit map.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Research adapter for localized provenance; choose license-compatible checkpoint explicitly.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermarkwm-wavmark
WavMark
Neural 32-bit watermark per one-second audio window; public wrapper reserves 16 bits for detection and 16 for user payload.
What it supports
Recovery of marked payload in the evaluated audio pipeline.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Public PyTorch implementation; documented mono 16 kHz input.
- Offline feasibility
- Yes, after obtaining dependencies and weights; local performance untested.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT repository; separately confirm redistributed model asset terms.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Authors test ten alterations and report over 38 dB SNR/4.3 PESQ; perceptual metrics are not universal inaudibility proof.
- False positives and false negatives
- The 1/2^16 pattern argument is per candidate under random-bit assumptions, not a calibrated file-level false-positive rate across many windows.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Audio baseline; report search-window multiplicity and calibrate real negatives.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermark / generator-integratedwm-gaussian-shading
Gaussian Shading
Distribution-preserving latent sampling encodes a message; diffusion inversion retrieves the latent and message.
What it supports
Expected latent watermark payload match under configured model/key.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Official PyTorch implementation includes encrypted randomization option.
- Offline feasibility
- Yes with model weights/key; GPU recommended.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code; base model licenses separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- “Performance-lossless” concerns the ideal latent distribution, not resistance to every edit or zero pixel distortion.
- False positives and false negatives
- Distribution and cryptographic assumptions matter; implementation options can weaken theoretical guarantees.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Research comparator for provenance at generation, not universal AI detection.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermark / generator-integratedwm-prc
Pseudorandom-code diffusion watermark (PRC)
Use pseudorandom error-correcting codes to select diffusion latents, combining cryptographic indistinguishability assumptions with watermark recovery.
What it supports
A matching keyed watermark is detected under stated model/security assumptions.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Official PRC-Watermark code; public example exposes binary detection; long-message interface described as forthcoming.
- Offline feasibility
- Yes with weights and key; diffusion inversion/generation compute required.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code; diffusion weights/licenses separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Paper reports robustness on Stable Diffusion 2.1 and 512-bit robust payload experiments; repository interface is narrower.
- False positives and false negatives
- Undetectability means an outsider cannot distinguish watermarking under assumptions; it does not mean a public detector can identify every AI image.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Security-oriented research comparator; distinguish theoretical properties, tested attacks and released features.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermark / generator-integratedwm-tree-ring
Tree-Ring watermark
Choose diffusion initial noise containing a structured Fourier-domain key; invert an image with a compatible model to test for the key.
What it supports
Image is statistically consistent with the configured latent watermark.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Official PyTorch/diffusers research implementation; version-sensitive DDIM inversion.
- Offline feasibility
- Yes with compatible diffusion weights and key; GPU/inversion compute significant.
- Cost and dependencies
- No quoted license fee; local compute and integration costs.
- Licenses and rights
- MIT code; generator/model/data licenses separate.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Paper evaluates geometric and photometric distortions; robustness depends on inversion and model matching.
- False positives and false negatives
- A statistical latent match is not cryptographic provenance or human identity.
- Privacy and data rights
- Prefer local processing; use opaque identifiers rather than personal data. Embedding does not establish rights.
- Potential LegitPix role
- Generator-integrated research family; do not describe as ordinary metadata fingerprinting.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
watermark / provenancewm-openai-provenance
OpenAI Content Provenance verifier (C2PA + SynthID)
Provider verifier reports separate provenance signals: image C2PA and SynthID; supported audio SynthID.
What it supports
Recognized OpenAI-origin watermark or OpenAI-issued AI-generation credentials, as separately reported.
What it does not prove
Scene truth, legal ownership, complete history, or absence of AI. A recovered mark is not itself a digital signature.
Implementation, limitations and sources
- Implementation and access
- Public web verifier and POST /v1/content_provenance_checks API; no top-level merged outcome in documented response.
- Offline feasibility
- No public offline verifier for SynthID confirmed; generic C2PA can be independently checked.
- Cost and dependencies
- API account/service dependencies; per-check pricing not verified.
- Licenses and rights
- Proprietary service under applicable OpenAI terms; no local media model license.
- Version or date
- 2026-10-06 review; source dates below
- Behavior under transformations
- Metadata may be lost while a watermark survives; degraded marks and legacy outputs may be missed.
- False positives and false negatives
- Not a general-purpose AI detector. C2PA not_detected can coexist with a valid third-party manifest; read validation_state and issuer independently.
- Privacy and data rights
- Uploads transmit media to OpenAI; require task-specific authorization. No upload made.
- Potential LegitPix role
- Separate provider-origin evidence adapter; preserve raw per-signal states, never reduce to real/fake.
- Uncertainty and open checks
- Research-only review; no media uploaded, model run or independent reproduction performed.
Primary sources
Evaluate before integrating
An authored test plan for LegitPix. These checks have not been run against the cataloged tools. Use deployment-specific measurements rather than transferring a paper's headline result.
State the decision first
Define whether the task is exact identity, near-copy retrieval, mark presence, payload attribution, signer validation, partial-edit localization, face manipulation, full synthesis or contextual verification. These require different ground truth and cannot share one accuracy label.
Build a rights-cleared test collection
Use media with documented permission, source and transformations. Separate real captures, ordinary edits, fully generated media, AI-assisted edits, composites and re-recordings. Include difficult legitimate cases. Keep subjects, capture devices, source videos and near-duplicates from leaking across train, validation and test splits.
Test the actual distribution path
Measure results on originals and ordinary deployment transformations: resizing, compression, format conversion, cropping, screenshots, metadata loss, video transcoding, frame-rate changes, excerpting, audio compression and recording conditions. Evaluate individual and combined transformations without presenting the results as removal instructions.
Report errors at the chosen operating point
Report true-positive rate at a predeclared false-positive rate, precision at plausible prevalence, false-negative rate, uncertainty intervals and abstention coverage. Include per-domain and low-quality-input results. AUROC, balanced-set accuracy and vendor headline numbers do not establish a suitable production threshold.
Evaluate the right unit
For matching, report candidate precision/recall at index scale and verify the reference link. For marks, separate present/absent detection, payload recovery, wrong-key or wrong-identity attribution, and perceptual quality. For provenance, test binding, signatures, trust decisions, time, revocation and missing ingredients independently. For localization, measure regions and time spans as well as whole-file classification.
Treat dependencies as evidence
Pin code, model weights, preprocessing, reference-index version and license. Measure CPU/GPU time, memory, file limits, failures and network calls on the intended platform. Test offline behavior explicitly. GPU inference demonstrated in a paper is not evidence that a model runs acceptably in a browser.
Validate updates before replacement
Maintain held-out, time-separated cases for new generators and devices. Track drift and regression, including false accusations on legitimate work. Correlated detectors are not independent witnesses; do not multiply their scores or average them into a probability without a validated fusion model.
Illustrative calculation, not a measured detector result: in 10,000 files with 1% target prevalence, 90% sensitivity and 1% false-positive rate yield about 90 true alerts and 99 false alerts. Only about 48% of alerts would be correct. The real prevalence, threshold and error rates must be measured for the deployment.
Measurement references
- NIST AI 100-4
Broad taxonomy and measurement guidance. Provenance and detection provide complementary evidence; low false-positive operating points and application context matter. The protocol in this library is an authored recommendation, not a NIST certification.
- NIST Open Media Forensics Challenge
Separates manipulation detection, deepfake detection and localization tasks. Documents ROC-based and localization metrics. The page says the evaluation has finished; archived benchmark results are not current deployment guarantees.
- Deepfake-Eval-2024
Image, audio and video benchmark of media circulating in 2024. Authors report substantial performance drops relative to earlier benchmarks. Its collection and labels do not establish every product's current accuracy or future generalization.
- NIST GenAI evaluation program
Current official program entry point links image, text and code challenge work. Use published task plans and current eligibility terms; no participation or dataset access was requested for this library.
- Watermarks and Fingerprints as Soft Bindings for Content Provenance
Original Pictures Technologies authors compare openly available watermark and fingerprint configurations under one protocol. Useful for evaluation design and payload-versus-presence distinctions; not peer-reviewed evidence of a universal winner. Findings need independent replication on the intended workload.
Scope, gaps and maintenance
Catalog schema, record IDs, CSV/JSON agreement, embedded data and search/filter/export logic were checked locally. Full browser rendering and actual browser downloads could not be verified in this environment; the HTML remains a self-contained reference with all records readable without JavaScript.
Baseline research completed 6 October 2026; bounded primary-source maintenance reviews completed 12, 19 and 26 October and 2 November 2026. Source-specific dates show which references were revisited. Implementations were not installed or benchmarked, commercial terms were not negotiated, and no personal media was uploaded. Recheck the exact version and licenses before adoption.
- Proprietary internals, closed detector keys, private reference databases, unpublished methods and contractual service terms cannot be fully inspected.
- Representative implementations cover the taxonomy; this is not an exhaustive bibliography or a universal performance leaderboard.
- No measured cross-method results were generated for this library. Robustness notes summarize cited evidence and must not be read as guarantees.
- Unverified release numbers, prices, license details and browser performance are marked explicitly rather than inferred.
- Dataset and media rights remain separate from a code license. Face/voice/device-linkage and public hash indexes may create privacy risks even without retaining original media.
- Text-only detection, document authorship, 3D assets, biometrics for identity decisions, moderation databases and legal ownership adjudication are outside the core image/audio/video scope.
This is a dated research snapshot. Before selecting a dependency, recheck the primary repository or official documentation, pin an immutable release/commit and model digest, resolve code/weights/dataset licenses separately, and run the evaluation protocol. This maintenance update did not activate accounts, signing services, plugins, subscriptions or public publishing.
Maintenance · 12 October 2026
Bounded review of standards, implementations, capture, watermarking, fingerprinting, detection, licenses and education. Existing record IDs, catalog content outside documented changes, and prior source verification dates are retained.
- capture-sony: Existing-documentation clarification, not newly released functionality: Avoid implying video support across every API or general availability for beta reporting/display tools.
- capture-truepic: Existing-documentation clarification, not newly released functionality: The existing broad SDK description can obscure materially different assurance boundaries.
- prov-cai-sdk: Document c2patool v0.28.2 and its external-manifest fix; distinguish Rust prerelease labeling from the separate C FFI Latest tag.
- prov-jpeg-trust: Update JPEG Trust Part 3 ballot milestone without describing the watermarking draft as a published standard.
- wm-synthid-google: Google announced SynthID Detector availability to everyone globally in English on October 7. The detector covers supported image, video and audio watermarks from Google and partners OpenAI, NVIDIA and Kakao; Apple is described as coming soon. Baseline early-tester-only and Google-origin-only wording is now stale.
- fd-sightengine-detection: October changelog adds image generator classes cosmos, hidream, hunyuan, krea, mai and vidu, and video classes cosmos, hailuo, happyhorse, hunyuan, krea, longlive, ltx and vidu. These families were previously classified under other. Baseline incorrectly identifies September as latest changelog month.
Sources checked and unresolved gaps
- https://authenticity.sony.net/camera/en-gb/ · 2026-10-12 · Latest release entry visible remains 2026.2, June 25, 2026. Paid upgrade/service and free default-certificate distinction remains.
- https://vision.truepic.dev/docs/compare-integration-approaches · 2026-10-12 · The existing broad SDK description can obscure materially different assurance boundaries. Vendor documentation. Do not interpret a server signature alone as hardware-attested sensor capture.
- https://www.truepic.com/lens-msa · 2026-10-12 · Lens MSA still describes proprietary server CLI/API/display products, and specifically treats CLI and display library as confidential. No open-source general detector or public unit price was established. Truepic news listing checked; newest displayed item was October 1, 2026, before baseline.
- https://www.truepic.com/blog · 2026-10-12 · Lens MSA still describes proprietary server CLI/API/display products, and specifically treats CLI and display library as confidential. No open-source general detector or public unit price was established. Truepic news listing checked; newest displayed item was October 1, 2026, before baseline.
- https://leica-camera.com/en-US/press/new-leica-m11-p · 2026-10-12 · M11-P remains a valid explicitly named hardware example. No edit necessary merely because the range has expanded. Optional context: Leica announced SL3-P availability and image Content Credentials on June 25, 2026, before baseline. Do not infer video signing or independent authenticity guarantees from that camera's video features.
- https://leica-camera.com/en-int/press/new-leica-sl3-p · 2026-10-12 · M11-P remains a valid explicitly named hardware example. No edit necessary merely because the range has expanded. Optional context: Leica announced SL3-P availability and image Content Credentials on June 25, 2026, before baseline. Do not infer video signing or independent authenticity guarantees from that camera's video features.
- https://leica-camera.com/en-GB/photography/content-credentials · 2026-10-12 · M11-P remains a valid explicitly named hardware example. No edit necessary merely because the range has expanded. Optional context: Leica announced SL3-P availability and image Content Credentials on June 25, 2026, before baseline. Do not infer video signing or independent authenticity guarantees from that camera's video features.
- https://www.iso.org/standard/77899.html · 2026-10-12 · ISO 24138:2024 and Apache-2.0 iscc-core still support baseline. Semantic-Code remains implemented but not algorithmically standardized by that edition. Opened core changelog starts with 1.3.0 dated March 2, 2026; no newer release is asserted from this cached page.
- https://github.com/iscc/iscc-core · 2026-10-12 · ISO 24138:2024 and Apache-2.0 iscc-core still support baseline. Semantic-Code remains implemented but not algorithmically standardized by that edition. Opened core changelog starts with 1.3.0 dated March 2, 2026; no newer release is asserted from this cached page.
- https://github.com/iscc/iscc-core/blob/main/CHANGELOG.md · 2026-10-12 · ISO 24138:2024 and Apache-2.0 iscc-core still support baseline. Semantic-Code remains implemented but not algorithmically standardized by that edition. Opened core changelog starts with 1.3.0 dated March 2, 2026; no newer release is asserted from this cached page.
- https://iscc.io/learn · 2026-10-12 · ISO 24138:2024 and Apache-2.0 iscc-core still support baseline. Semantic-Code remains implemented but not algorithmically standardized by that edition. Opened core changelog starts with 1.3.0 dated March 2, 2026; no newer release is asserted from this cached page.
- https://github.com/facebook/ThreatExchange/tree/main/pdq · 2026-10-12 · Public PDQ implementation remains in ThreatExchange. Repository license still uses three BSD conditions; README retains file-level exception warning, including CImg. No algorithm or license change verified.
- https://github.com/facebook/ThreatExchange · 2026-10-12 · Official vPDQ comparison still distinguishes TMK fixed-length whole-video/same-length matching from vPDQ clip matching. No record change needed; TMK source code itself was not re-inspected.
- https://github.com/facebook/ThreatExchange/blob/main/LICENSE · 2026-10-12 · Official README still states wheels are not distributed. Variable-length fingerprints and subsequence/clip matching distinction remain; inherited PDQ limitations and candidate-filter false negatives are documented.
- https://github.com/facebook/ThreatExchange/blob/main/vpdq/README.md · 2026-10-12 · Official vPDQ comparison still distinguishes TMK fixed-length whole-video/same-length matching from vPDQ clip matching. No record change needed; TMK source code itself was not re-inspected.
- https://github.com/facebookresearch/sscd-copy-detection · 2026-10-12 · Official repository remains the CVPR 2022 research implementation and explicitly describes its codebase as MIT licensed. Separate model-weight and dependency clearance caveats should remain; no independent deployment accuracy or maintained SLA established.
- https://contentauthenticity.org/how-it-works · 2026-10-12 · Existing education resource, newly linked; republication rights were not verified.
- https://contentauthenticity.org/media-literacy · 2026-10-12 · Existing education resource, newly linked; republication rights were not verified.
- https://learn.contentauthenticity.org/ · 2026-10-12 · Existing education resource, newly linked; republication rights were not verified.
- https://spec.c2pa.org/specifications/specifications/2.4/index.html · 2026-10-12 · C2PA 2.4 remains the top published navigation version; no newer release verified.
- https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html · 2026-10-12 · Published 2.4 technical specification retained.
- https://spec.c2pa.org/specifications/specifications/2.4/softbinding/Decoupled.html · 2026-10-12 · Soft-binding architecture retained; no protocol change established.
- https://cawg.io/identity/1.3/ · 2026-10-12 · 1.3 remains DIF-ratified 17 August 2026.
- https://www.iso.org/standard/91405.html · 2026-10-12 · Part 1 edition 2 remains published August 2026.
- https://www.iso.org/standard/89038.html · 2026-10-12 · Part 2 remains FDIS at 50.00.
- https://www.iso.org/standard/90209.html · 2026-10-12 · Part 3 moved to 40.60 on 7 October; still a draft.
- https://www.iso.org/standard/93391.html · 2026-10-12 · Part 4 still shown as DIS at 40.20; cached page may lag.
- https://github.com/contentauth/c2pa-rs · 2026-10-12 · Current README identifies the SDK as beta and distinguishes active development from released versions.
- https://github.com/contentauth/c2patool · 2026-10-12 · Separate tool repository and MIT/Apache licensing retained.
- https://contentauth.github.io/c2pa-js/ · 2026-10-12 · Documentation available; no new package-version conclusion drawn.
- https://github.com/contentauth/c2pa-js · 2026-10-12 · MIT license indicated; no package release or compatibility guarantee inferred.
- https://github.com/contentauth/c2patool/releases/tag/v0.28.2 · 2026-10-12 · Latest release 8 October; SDK dependency and external-manifest fixes.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-v0.91.2 · 2026-10-12 · 6 October tag explicitly labeled Pre-release; parser/manifest hardening in notes.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-c-ffi-v0.91.2 · 2026-10-12 · C FFI tag separately labeled Latest; release-body date differs from page posting date.
- https://c2pa.org/conformance/ · 2026-10-12 · Operational lists and frozen legacy ITL guidance retained.
- https://c2pa.org/faqs/ · 2026-10-12 · Certificate onboarding costs and legacy trust-model distinctions retained.
- https://spec.c2pa.org/conformance-explorer/ · 2026-10-12 · Dynamic page returned no readable entries; no individual product conformance newly claimed.
- https://blog.google/innovation-and-ai/models-and-research/google-deepmind/synth-id-ai-content/ · 2026-10-12 · Google announced SynthID Detector availability to everyone globally in English on October 7. The detector covers supported image, video and audio watermarks from Google and partners OpenAI, NVIDIA and Kakao; Apple is described as coming soon. Baseline early-tester-only and Google-origin-only wording is now stale.
- https://deepmind.google/models/synthid/ · 2026-10-12 · Google announced SynthID Detector availability to everyone globally in English on October 7. The detector covers supported image, video and audio watermarks from Google and partners OpenAI, NVIDIA and Kakao; Apple is described as coming soon. Baseline early-tester-only and Google-origin-only wording is now stale.
- https://sightengine.com/docs/changelog · 2026-10-12 · October changelog adds image generator classes cosmos, hidream, hunyuan, krea, mai and vidu, and video classes cosmos, hailuo, happyhorse, hunyuan, krea, longlive, ltx and vidu. These families were previously classified under other. Baseline incorrectly identifies September as latest changelog month.
- https://github.com/facebookresearch/videoseal · 2026-10-12 · Current README retains MIT code/models, stable 256-bit VideoSeal, 256-bit PixelSeal and 1024-bit ChunkySeal; latest listed model release remains December 2025. No new post-baseline release established.
- https://github.com/facebookresearch/audioseal · 2026-10-12 · MIT code/weights and streaming support confirmed. Root README and CHANGELOG disagree on the 0.2 date (December 2024 versus December 2025), so do not add a precise version release date from either alone.
- https://github.com/facebookresearch/watermark-anything · 2026-10-12 · Archived August 1, 2026 remains confirmed; new SA-1B model/code MIT, original COCO model noncommercial, as baseline already states.
- https://github.com/facebookresearch/stable_signature · 2026-10-12 · Predominantly CC-BY-NC, with MIT ldm/taming subtrees; no license change established.
- https://arxiv.org/abs/2610.04151 · 2026-10-12 · Soft-binding benchmark remains v1 submitted October 2, 2026; expected-payload versus ECC-only caveat is already present. It is a single preprint, not a new release or independent reproduction by this review.
- https://github.com/Amamiya-C/PROBE-AIGI-Detection · 2026-10-12 · Repository still has unchecked evaluation/weights, detector-finetuning and boundary-exploration TODOs alongside evaluation code/checkpoint links. No top-level license visible. Retain baseline's partial-release caution.
- https://github.com/yermandy/GenD · 2026-10-12 · MIT repository and released pretrained backbone variants remain documented; no new post-baseline change established.
- https://github.com/YZY-stack/Effort-AIGI-Detection · 2026-10-12 · Face and general-image checkpoint split remains documented; no new top-level license grant established.
- https://docs.thehive.ai/docs/ai-image-and-video-detection · 2026-10-12 · Visual source classes, separate deepfake scores, sampled-frame rules and use-case threshold refinement remain documented. No dated post-baseline change established.
- https://docs.thehive.ai/docs/ai-generated-audio-detection · 2026-10-12 · Current docs describe per-10-second audio classification; no dated new release found. Provider scores remain unvalidated.
- GitHub API release/commit endpoints were inaccessible via the web tool. Main-branch pages were observed, but commit hashes and complete October 6–12 change histories were not pinned. Do not claim exhaustive absence of changes.
- Some web pages were returned from recent cached crawls. Access date does not establish their publication date.
- No software installed, model executed, device firmware tested, certificate chain validated, account opened, contract accepted, media uploaded, or service price negotiated.
- Public educational material was not fully licensed for republication; linking is the proposed use.
- Leica current range includes additional models, but full model/firmware-level capture-signing capabilities were outside this bounded review.
- This is a bounded maintenance sweep, not a re-verification of every catalog source. Unchecked source dates remain unchanged.
- No implementations, model weights or media were installed, benchmarked or uploaded. Vendor performance remains vendor-reported.
- The dynamic Conformance Explorer did not expose readable entries; individual products were not newly certified by this review.
- Full ISO standards were not acquired; status statements rely on public ISO listings, with cache freshness limitations.
- A targeted public-source review is not an exhaustive scan of all new research or every commit.
- Google announcement verified public release date; availability was not tested with login or media.
- Sightengine changelog only states a month, not a day.
- Some GitHub/registry pages returned stale cached snapshots or failed. Conflicting observations are retained rather than resolved by guessing.
- No implementation, license dependency, privacy contract or model performance audit beyond public documentation was performed.
- Speech-detection access documentation could not be fetched during this run; the earlier beta/access caveat remains, with its prior review date.
- Some optional implementation-limit clarifications were identified but left outside this bounded update; no new method was added from a recent unreplicated preprint.
Maintenance: 19 October 2026
- prov-cai-sdk: Record October 16 SDK and C FFI 0.91.3 releases, preserving their different release labels and untested status.
- prov-jpeg-trust: Correct Part 4 from stale 40.20 to dated ISO DIS 40.60; voting closed September 18, not a newly published standard.
Checked sources and unresolved gaps
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-v0.91.3: New October 16 Rust release, explicitly prerelease; maintainer release notes, not independent testing. Checked 2026-10-19.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-c-ffi-v0.91.3: Separate October 16 C FFI release is Latest. Checked 2026-10-19.
- https://docs.rs/crate/c2pa/latest: Published crate 0.91.3 corroborated. Checked 2026-10-19.
- https://github.com/contentauth/c2patool/releases: 0.28.2 remains latest visible release. Checked 2026-10-19.
- https://docs.rs/crate/c2patool/latest: Package baseline reconfirmed. Checked 2026-10-19.
- https://spec.c2pa.org/specifications/specifications/2.4/index.html: Highest listed specification 2.4; no newer ratification found. Checked 2026-10-19.
- https://cawg.io/identity/1.3/: Version 1.3 ratified August 17 retained. Checked 2026-10-19.
- https://cawg.io/: Other branches remain drafts. Checked 2026-10-19.
- https://www.iso.org/standard/91405.html?browse=tc: Part 1 edition 2 published August 26. Checked 2026-10-19.
- https://www.iso.org/standard/89038.html?browse=tc: Part 2 FDIS 50.00 remains under development. Checked 2026-10-19.
- https://www.iso.org/standard/90209.html: Part 3 DIS 40.60 remains under development. Checked 2026-10-19.
- https://www.iso.org/standard/93391.html?browse=tc: Part 4 DIS 40.60 September 18; correction of older cached baseline. Checked 2026-10-19.
- https://jpeg.org/jpegtrust/workplan.html: Stale stages conflict with dated ISO records. Checked 2026-10-19.
- https://sightengine.com/docs/changelog: Visible October class additions already recorded; no newer meaningful change established. Checked 2026-10-19.
- https://github.com/facebookresearch/videoseal: Visible model release baseline unchanged; MIT code/models documented by maintainers. Checked 2026-10-19.
- https://deepmind.google/models/synthid/: Public supported-partner Detector access remains documented; Apple still described as coming soon. Checked 2026-10-19.
- https://authenticity.sony.net/camera/en-gb/: Latest listed solution release June 25; image/video and beta boundaries unchanged. Checked 2026-10-19.
- https://github.com/facebook/ThreatExchange/tree/main/pdq: Perceptual matching and implementation-dependent hash cautions retained; page cached last month. Checked 2026-10-19.
- https://contentauthenticity.org/media-literacy: Education reference remains available; republication license still unverified. Checked 2026-10-19.
- https://github.com/iscc/iscc-core/releases: Cached release listing insufficient to establish newer version; existing baseline retained. Checked 2026-10-19.
- Bounded review; sources not rechecked retain their earlier dates. Some web results are cached, especially GitHub and ISO variants.
- Release fixes are maintainer-reported and were not installed or independently tested; no conformance or performance certification is implied.
- No full ISO standard, source article, model, dataset or personal media was copied. Public education reuse rights and transitive licensing remain to be checked before adoption.
- No new commercial pricing, contracts, independent detector evaluation or device-level capture validation established.
- Full visual browser rendering and actual browser download behavior remain unverified.
Maintenance: 26 October 2026
- prov-c2pa-hard: Record official C2PA 2.5 October specification; distinguish specification features from implemented support.
- prov-cai-sdk: Update to c2patool 0.28.3 and SDK/C FFI 0.91.6; record intervening timestamp and MOV fixes with component-specific release labels.
- prov-jpeg-trust: Resolve prior JPEG workplan disagreement: Part 1 is published and Part 4 remains DIS 40.60, not published.
Checked sources and unresolved gaps
- https://spec.c2pa.org/specifications/specifications/2.4/index.html: Navigation now points to 2.5, superseding last review’s highest-listed 2.4 observation. Checked 2026-10-26.
- https://spec.c2pa.org/specifications/specifications/2.5/index.html: Official 2.5 specification set available; publication is not universal implementation support. Checked 2026-10-26.
- https://spec.c2pa.org/specifications/specifications/2.5/specs/C2PA_Specification.html: October 2026 version history verified; expanded formats/live media, linked-resource integrity, ML-DSA and validation changes. Checked 2026-10-26.
- https://github.com/contentauth/c2pa-rs/releases: Rust 0.91.6 Pre-release and C FFI 0.91.6 Latest released October 24. C FFI body retains October 16 notes; do not infer a new label-setting feature from that body. Checked 2026-10-26.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-v0.91.4: October 19 maintainer release changes SHA-1 timestamp-imprint handling. Checked 2026-10-26.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-v0.91.5: October 23 maintainer release fixes valid MOV rejection. Checked 2026-10-26.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-v0.91.6: October 24 release updates cryptographic dependencies; not independently tested. Checked 2026-10-26.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-c-ffi-v0.91.6: Direct page cache miss; release identity/label/date confirmed from official aggregate listing instead. Checked 2026-10-26.
- https://github.com/contentauth/c2patool/releases: Latest listed version now 0.28.3. Checked 2026-10-26.
- https://github.com/contentauth/c2patool/releases/tag/v0.28.3: October 24 release adopts SDK 0.91.6. Checked 2026-10-26.
- https://docs.rs/crate/c2pa/latest: Published 0.91.6 confirmed. Checked 2026-10-26.
- https://github.com/contentauth/c2pa-rs/blob/main/LICENSE-MIT: MIT license remains present; complete transitive-license audit not performed. Checked 2026-10-26.
- https://cawg.io/: Current identity remains 1.3; draft branches remain separately labeled. Checked 2026-10-26.
- https://cawg.io/identity/1.3/: Identity 1.3 current; no newer identity ratification established. Checked 2026-10-26.
- https://jpeg.org/jpegtrust/workplan.html: Now agrees with published Part 1 edition 2 and DIS Part 4 stage 40.60; Parts 2-4 still under development. Checked 2026-10-26.
- https://deepmind.google/models/synthid/: Public Detector and supported partners remain documented; Apple still coming soon. Vendor claims, not independent evaluation. Checked 2026-10-26.
- https://github.com/facebookresearch/videoseal: Latest listed model announcements remain December 2025; maintainers document MIT code/models. Checked 2026-10-26.
- https://github.com/facebook/ThreatExchange/tree/main/pdq: Reference implementations and platform-dependent hashing cautions retained; cached last month, so no fresh release conclusion. Checked 2026-10-26.
- https://sightengine.com/docs/changelog: October generator-class changes already recorded; no newer meaningful change established. Checked 2026-10-26.
- https://authenticity.sony.net/camera/en-gb/: Latest listed solution release June 25; no new release established. Device/service claims remain vendor-reported. Checked 2026-10-26.
- https://contentauthenticity.org/media-literacy: Education curricula available; independent republication rights not established. Checked 2026-10-26.
- https://mfc.nist.gov/: Fetch timed out; no current benchmark conclusion drawn. Checked 2026-10-26.
- Bounded review; unchanged records retain earlier review dates. Cached source pages cannot establish absence of newer releases.
- Exact C2PA 2.5 publication day, per-implementation 2.5 coverage, conformance, security and performance were not independently evaluated.
- C FFI direct release page cache miss; aggregate official listing corroborates date and label. NIST MFC fetch timed out.
- Commercial pricing/contracts, transitive licenses, education republication rights and independent detector/capture validation remain open.
- Full browser rendering and actual download behavior remain unverified; static structure and embedded/offline payloads checked.
Maintenance: 2 November 2026
Bounded primary-source review across provenance, signatures, watermarking, fingerprinting, detection, capture, standards, implementations, licensing and education. Existing IDs and unrelated user content retained; no media uploaded or software benchmarked.
- prov-cai-sdk: Record October 26 evening c2patool 0.28.4 and Rust/C FFI 0.91.7 releases and TLS dependency maintenance; keep component-specific labels, cached crate-page uncertainty, and untested status.
- prov-jpeg-trust: Label earlier ISO/JPEG workplan discrepancies explicitly as historical; current published-versus-draft status is unchanged.
Checked sources and unresolved gaps
- https://github.com/contentauth/c2pa-rs/releases: Rust 0.91.7 Pre-release and C FFI 0.91.7 Latest released October 26 evening, after the previous review; no later release established from this listing. Checked 2026-11-02.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-v0.91.7: Maintainer notes identify TLS dependency minimums and eMudhra distrust updates; implementation not installed or tested. Checked 2026-11-02.
- https://github.com/contentauth/c2pa-rs/releases/tag/c2pa-c-ffi-v0.91.7: Direct page corroborates October 26 date and Latest label; body repeats October 16 API notes, not newly released functionality. Checked 2026-11-02.
- https://github.com/contentauth/c2patool/releases: Latest visible version is 0.28.4. Checked 2026-11-02.
- https://github.com/contentauth/c2patool/releases/tag/v0.28.4: October 26 release adopts SDK 0.91.7; maintainer-reported dependency update. Checked 2026-11-02.
- https://docs.rs/crate/c2pa/latest: Returned cached 0.91.6 page, so not used to assert 0.91.7 crate publication or absence. Checked 2026-11-02.
- https://docs.rs/crate/c2pa/0.91.7: Fetch inaccessible; publication corroboration remains open, not a finding that the crate is absent. Checked 2026-11-02.
- https://github.com/contentauth/c2pa-rs/blob/main/LICENSE-MIT: MIT license file remains available. No transitive-license or legal audit performed. Checked 2026-11-02.
- https://spec.c2pa.org/specifications/specifications/2.5/index.html: Official navigation still lists 2.5 as current. Specification publication does not establish feature support in every implementation. Checked 2026-11-02.
- https://jpeg.org/jpegtrust/workplan.html: Published Part 1 edition 2; Part 2 FDIS 50.00 and Parts 3/4 DIS 40.60 remain under development. Earlier disagreement is historical, not current. Checked 2026-11-02.
- https://cawg.io/: Identity 1.3 remains current; governance, vc-vp and vlei variants are explicitly drafts. No new ratification established. Checked 2026-11-02.
- https://deepmind.google/models/synthid/: Public SynthID Detector and supported partner coverage remain documented; Apple still described as coming soon. Provider claims, not independent evaluation. Checked 2026-11-02.
- https://github.com/facebookresearch/videoseal: Latest model announcements shown remain December 2025; maintainers state MIT for models/code. Robustness assertions are author claims, not new independent validation. Checked 2026-11-02.
- https://github.com/facebook/ThreatExchange/tree/main/pdq: Reference implementations and platform-dependent decoding/hash variation cautions remain. Cached last-month page cannot establish current release absence. Checked 2026-11-02.
- https://sightengine.com/docs/changelog: October generator-class additions already recorded; no newer listed change established. Changelog explicitly excludes some internal accuracy/performance updates; no independent performance claim inferred. Checked 2026-11-02.
- https://authenticity.sony.net/camera/en-gb/: Newest listed solution release remains June 25, 2026. Beta labels, API still-image scope, and paid service/free certificate distinction remain; vendor assurance claims not independently tested. Checked 2026-11-02.
- https://vision.truepic.dev/docs/compare-integration-approaches: Native apps list device attestation and offline capture; browser/WebView variants do not. Existing assurance-boundary caveats retained; vendor documentation only. Checked 2026-11-02.
- https://contentauthenticity.org/media-literacy: Education curriculum resource still available. General permission for independent republication not established; no source curriculum copied. Checked 2026-11-02.
- https://www.nist.gov/programs-projects/generative-artificial-intelligence-evaluation-program-genai: Program remains ongoing and links image 2025/2026 challenge; page update May 7. No new published benchmark results or certification inferred. Checked 2026-11-02.
- https://arxiv.org/abs/2503.02857v5: Pinned v5 remains May 27, 2026. Cross-domain performance findings concern the authors tested models/data, not every current detector or a universal real/fake verdict. Checked 2026-11-02.
Unresolved gaps
- Bounded review, not an exhaustive search; unchanged method records retain their prior review dates. Cached pages cannot establish that no newer release exists.
- The docs.rs latest page returned cached 0.91.6 and pinned 0.91.7 fetch failed. GitHub release metadata is verified; independent crate-publication corroboration remains open.
- No independent security, C2PA 2.5 conformance, detector-performance or capture-assurance tests were run. Vendor assertions are not independent evaluations.
- Commercial terms, transitive code/weights/data licenses and education republication rights remain to be checked before adoption. No accounts, signing services, plugins, sponsorships, public publishing or access changes were activated.
- Full browser rendering and real download interactions remain unverified: local Chromium launch was blocked by the environment (socket permission). No attempt was made to bypass that restriction. Static structure, embedded payloads and local JavaScript logic were validated separately.
Primary source index
Linked references are grouped within the method records above. This index provides a quick route to the original documents; access to the originals requires an internet connection. No source articles, model weights or datasets are bundled.
Browse 193 source links
- A leak in PRNU based source identification: Questioning fingerprint uniqueness
- A Self-Supervised Descriptor for Image Copy Detection
- AASIST ASVspoof5 baseline
- AASIST official implementation
- AcoustID commercial service
- AcoustID data and software license
- AcoustID web service documentation
- ACRCloud custom-file results and project types
- ACRCloud identification API
- An undetectable watermark for generative image models
- ASVspoof5 Phase 2 evaluation plan
- ATSC A/334 current standard
- ATSC A/335 current standard
- audfprint author implementation
- AudioSeal repository
- audiowmark detailed documentation
- audiowmark repository
- Automatic Detection of Internal Copy-Move Forgeries in Images
- AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection
- Bender et al., Techniques for data hiding
- c2pa C FFI v0.91.2 release page
- c2pa C FFI v0.91.3 release page
- c2pa C FFI v0.91.7 release page
- C2PA capture/scene security limits
- C2PA Certificate Policy
- c2pa component release listing
- C2PA Conformance Explorer
- C2PA explainer
- C2PA FAQ
- C2PA Soft Binding API
- C2PA technical specification
- C2PA technical specification 2.5
- c2pa-js documentation
- c2pa-rs
- c2pa-v0.91.2 release notes
- c2pa-v0.91.3 release notes
- c2pa-v0.91.4 release notes
- c2pa-v0.91.5 release notes
- c2pa-v0.91.6 release notes
- c2pa-v0.91.7 release notes
- c2patool
- c2patool v0.28.2 release notes
- c2patool v0.28.3 release notes
- c2patool v0.28.4 release notes
- CAI media literacy curriculum
- CAI public explanation and FAQ
- CAWG Identity Assertion
- Celik et al., hierarchical fragile authentication watermarking
- Chen–Wornell, Quantization Index Modulation
- Chromaprint complete licensing statement
- Chromaprint official README
- ChunkySeal model card
- Client license
- CLIP official model card
- CLIP official repository
- CNNDetection author implementation
- CNNDetection license
- Content Credentials Foundations
- Cosign repository
- Cosign verification
- Cox et al., Secure Spread Spectrum Watermarking for Multimedia
- Deepfake Detection that Generalizes Across Benchmarks
- Deepfake-Eval-2024
- DeepfakeBench official framework and license
- Digimarc Content Credentials extension
- Digimarc product description
- Digimarc Validate product overview
- DIRE for Diffusion-Generated Image Detection
- DIRE official implementation
- Effort author implementation
- ExifTool application documentation
- ExifTool official README and license
- Exposing Digital Forgeries by Detecting Inconsistencies in Lighting
- Exposing Digital Forgeries in Complex Lighting Environments
- FFmpeg formats: hash, framehash and streamhash muxers
- FFmpeg license and legal considerations
- Forensically official tool documentation
- Gaussian Shading official repository
- Gaussian Shading paper
- GenD official implementation
- Google announces global public SynthID Detector
- Google Search: Search with an image
- Gruhl, Lu and Bender, Echo hiding
- HiDDeN paper
- Hive audio detection
- Hive FAQ: retention and operating assumptions
- Hive image and video detection
- How YouTube Content ID works
- Image Forgeries Detection through Mosaic Analysis
- ImageHash official repository
- ImageHash PyPI release history
- IMATAG API v4
- IMATAG payload documentation
- IMATAG pricing
- invisible-watermark official repository
- IPTC AI metadata additions
- IPTC Photo Metadata Standard
- IPTC synthetic-media guidance
- ISCC core reference implementation
- ISCC explanation and semantic-code status
- ISCC standard scope and structure
- ISO 24138:2024 official abstract
- ISO edition 2 status
- JPEG Trust overview
- JPEG Trust profiles and reports draft status
- JPEG Trust reference software draft status
- JPEG Trust reference software updated ISO lifecycle view
- JPEG Trust watermarking draft status
- JPEG Trust workplan
- Leica M11-P official launch
- Lin–Chang, Semi-Fragile Watermarking for Authenticating JPEG Visual Content
- Local JPEG Grid Detector via Blocking Artifacts
- Meta ThreatExchange PDQ
- Meta ThreatExchange project overview and licensing
- Microsoft PhotoDNA official overview
- NAGRA Akamai integration
- NexGuard Streaming product sheet page
- NIST AI 100-4
- NIST FIPS 180-4: Secure Hash Standard
- NIST GenAI evaluation program
- NIST Open Media Forensics Challenge
- Noiseprint author implementation and license
- NoiseSniffer author implementation
- Official HiDDeN repository
- OpenAI advancing content provenance
- OpenAI Content Provenance API guide
- OpenCV feature matching and homography
- OpenCV licensing
- OpenTimestamps client
- OpenTimestamps official site
- Original checkpoint license
- Orthogonal Subspace Decomposition for Generalizable AI-Generated Image Detection
- Panako author implementation
- Pixel Seal paper
- PixelSeal model card
- Politecnico di Milano PRNU Python implementation
- PRC-Watermark repository
- Primary paper university-hosted copy
- PROBE official implementation
- Published c2pa crate
- Qualify for Content ID
- Reality Defender API quickstart
- Reality Defender RealAPI: scope, plans and deployment
- RFC 3161
- RivaGAN official repository
- SARI
- Sightengine AI content detection model index
- Sightengine API changelog
- Sightengine privacy policy, subscriber-data section
- Sightengine speech detection beta/access documentation
- Sigstore overview
- SilentCipher paper
- SilentCipher repository
- Soft-binding benchmark
- Sony Camera Authenticity Solution
- Sony video service announcement
- SSCD official implementation
- Stable Signature official repository
- StegaStamp official repository
- SWGDE Best Practices for Digital Audio Authentication
- SWGDE Best Practices for Digital Video Authentication
- SWGDE Best Practices for Image Authentication
- SWGDE Electric Network Frequency Discussion Paper
- SynthID official product page
- SynthID-Image technical report
- TALL and TALL++ author repository
- TALL: Thumbnail Layout for Deepfake Video Detection
- Tian, Reversible Data Embedding Using a Difference Expansion
- TinEye MatchEngine private image search
- TinEye: How does TinEye work?
- TinEye: What is TinEye?
- TMK+PDQF official implementation
- Toward secure public-key blockwise fragile authentication
- Tree-Ring official repository
- Tree-Ring paper
- Truepic controlled capture SDK announcement
- Truepic current integration comparison
- Truepic Lens service agreement
- TruFor author implementation and license
- TrustMark C2PA integration
- TrustMark repository documentation
- UniversalFakeDetect official repository
- Video Seal paper
- VideoSeal repository: license and availability
- VideoSeal v1.0 model card
- vPDQ README and comparison with TMK
- Watermark Anything repository
- Watermarks and Fingerprints as Soft Bindings for Content Provenance
- WavMark paper
- WavMark repository
- We Can Hide More Bits
- Where Detectors Fail: Probing Generative Space for Generalizable AI-Generated Image Detection
- ZERO: Local JPEG Grid Origin Detector