Transparent methods · Local by default
Signals, scores, and limits
GenDetect defaults to local provenance and supporting-evidence checks. When configured, full analysis uses open models and specialist services. These models are not yet calibrated on our business data; a screening index is not an AI probability, and insufficient evidence still requires human review.
01 · Implemented signals
Each input type follows a different browser analysis path. This page describes signals the code actually reads and does not present unimplemented model capabilities as product features.
After normalizing whitespace, rules compare vocabulary diversity, repeated three-token sequences, average sentence length, punctuation frequency, and unusual-character rate. Fewer than 40 analyzable tokens returns insufficient evidence and no screening index.
Open the AI text detectorThe browser first checks file bytes for known generator and algorithmic-media declarations, then uses Canvas to sample color buckets, luminance variance, and adjacent-pixel edge density. Metadata remains an unverified clue. The local flow validates a C2PA signature chain; only a trusted, asset-bound explicit claim supports a source conclusion. Configured full analysis can use Community Forensics and specialist review.
Open the AI image detectorThe browser samples static frames near 20%, 50%, and 80% of the video, keeps pixel rules as supporting clues, and displays duration, resolution, estimated bitrate, extension, and MIME. It does not inspect audio or full temporal continuity.
Open the AI video detector02 · Score meaning
The new interface does not combine text, image or frame rules into one score. Each model output is shown independently; it is not a calibrated probability or statistical confidence interval.
High repetition, limited color diversity and unusual edges are supporting heuristic clues only. Metadata, provenance verification and model outputs are not summed or averaged into an AI probability.
Text with fewer than 40 analyzable tokens returns insufficient evidence. Images below 128 pixels on either edge also abstain unless a generator metadata claim is present. If video frames cannot be decoded, the result falls back to metadata facts only.
The current result cannot identify a generation model, author, or editing process. Similar statistical patterns may appear in templated human writing, compressed media, and AI-generated content.
03 · Known limitations
Any detector that sees only the final artifact faces lost information and distribution change. The cases below can create both false positives and false negatives.
Human rewriting, translation, screenshots, filters, crops, platform compression, and repeated transcoding all alter original signals. They may hide generation traces or create patterns that look unusual.
Generation systems keep changing, and people can deliberately adjust rhythm, noise, or encoding parameters. Fixed rules cannot cover generation methods that have not yet been observed.
Code, poetry, tables, brand templates, low-light photos, and illustrations may naturally cross text or image thresholds. Unusually encoded video only produces metadata warnings and never adds a generation score. A high index is not proof, and a low index does not prove human authorship.
04 · Verification and privacy
Reliable conclusions come from provenance, timing, and editing history, not one score. The current implementation keeps input on the device and commits to disclosing future capability changes.
Preserve complete text, the original file, publication time, author information, and edit history. Screenshots and reposts discard provenance signals that could support verification.
Use content credentials, reverse-image search, trusted-source versions, audiovisual continuity, and contextual facts. Important decisions require human review and a chance for the person being assessed to explain.
Text, video, and the default image heuristic run in the browser. An image is uploaded only after the user selects SynthID, SD/Flux, or an available external Verify mode. External Verify may also forward it to the provider named in the interface; the interface and privacy notice state purpose, retention, and deletion.