Image
Decode JPEG or PNG input, recognize repeat assets, then score the image. Returns a warning label, raw score, active model, thresholds and cache state.
POST /v1/assets/checkHow it works
From content
to context.
Detection and decision remain separate. Every result can be inspected, challenged and overridden.
Open API reference- 01
Submit
Send decoded media or client-side fingerprints.
- 02
Recognize
Exact hashes, perceptual hashes and stable footprints find repeats.
- 03
Analyze
The active modality model returns a raw warning signal.
- 04
Contextualize
Attach version, limits, provenance and quality reasons.
- 05
Decide
A reviewer records disclosure or publication action.
Sampled video
Submit 5–8 representative frames and a stable video footprint. The API aggregates frame signals. It is not a temporal deepfake model and does not evaluate motion continuity or lip-sync.
POST /v1/videos/check-framesUnknown is information.
No usable input, no cache match or no safe model answer should not be forced into real or fake.
Every useful response should expose
- Modality and status
- Warning label and raw model score
- Model name and version
- Quality reason and limitations
- Cache or storage source
- Recommended human action
The review dossier adds what a detector cannot.
Who supplied it and what tools were declared.
Content Credentials present, missing or invalid.
Lane, model, score, warning and limitations.
Disclosure or publication choice with rationale.