Informational output
Metadata values can be incomplete, edited, forged, removed, compressed, or stored in unsupported application-specific structures.
Audio and video records
Media tags can be absent, edited, forged, stripped, or stored in unsupported private structures. Codec, duration, frame-rate, and stream information describe only the inspected container and do not establish authorship, licensing, originality, or whether content was AI-generated.
Visual context
The separate visual estimate is probabilistic and can produce false positives and false negatives. It does not validate metadata, reconstruct missing fields, identify a specific generator, or prove human or AI authorship.
Generated captions
Captioning is optional and runs locally with a same-origin model, using WebGPU when available. Suggested titles, descriptions, and tags may hallucinate or omit content and must be reviewed. Video results use sampled frames, not the complete stream. Caption output remains local and is not added to automatic public UUID results in this release. The feature must not be used to infer identity, exact age, or ethnicity.
Visual inventory
Object regions and COCO labels are probabilistic, vocabulary-limited suggestions. They may omit visible subjects, mislabel content, or reflect model bias. They are not metadata, accessibility descriptions, face recognition, identity findings, safety classifications, or proven probabilities that an object is present.
N-rated content estimate
The N-rated result is a probabilistic screening signal for visible nudity or sexual content, not an age rating, legal judgment, or finding about any person’s age, identity, consent, or intent. It can produce false positives and false negatives, including for artistic, medical, breastfeeding, fitness, swimwear, cropped, compressed, and low-context images. Do not use it as the sole basis for accusations, sanctions, or legal conclusions.
Automatic public image results
Choosing a still image starts analysis and publication in one disclosed action. A sanitized preview and normalized report, including published text fields and model outputs, are listed publicly for up to 24 hours. You must have the right to publish them and must not submit unlawful, non-consensual intimate, confidential, personally sensitive, or rights-infringing content. Audio and video remain local-only. The public UUID preview is replaced with a neutral placeholder and the result is excluded from Recent results when the N-rated estimate is uncertain, detected, or unavailable; this screening is fallible and is not a guarantee that restricted content will be caught. Deletion or expiry cannot revoke copies already saved or cached by other services.
User responsibility
You remain responsible for how extracted prompts, workflows, models, and other records are used and shared.
No recovery guarantee
The tool cannot recover information that is absent from the supported containers, and it does not reconstruct prompts from image appearance.
Coverage limits
The tool does not decrypt protected media, repair damaged files, identify music, transcribe speech, detect synthetic audio, analyze every video frame, or verify proprietary audio/video watermarks. An unsupported or incomplete result does not prove metadata is absent.
Changing formats
Coverage may change as file formats, C2PA specifications, browser capabilities, visual models, and generator conventions evolve.