Scan a video for unsafe or policy-violating content. Supports MP4, MOV, and WebM URLs and returns structured safety classifications across harassment, hate, sexual, sexual-minors, and violence categories — useful for moderating user-generated or AI-generated video before publishing.
About this model
Scan a video for unsafe content across the major safety categories — harassment, hate, sexual, sexual-involving-minors, and violence. Submit a video URL and an optional context prompt, and receive a structured verdict you can use to gate publishing or trigger human review.
Cost analysis
| Provider | Cost | Notes |
|---|---|---|
| muapiapp | $0.02 per video | Flat fee per moderation call regardless of duration. |
| Fal.ai | Not available | No equivalent first-party video-moderation model. |
| Replicate | Not available | No equivalent first-party video-moderation model. |
Flat fee per moderation call regardless of duration.
No equivalent first-party video-moderation model.
No equivalent first-party video-moderation model.
* Competitor pricing is estimated based on similar model architectures and usage tiers.
Configuration schema
| Parameter | Type | Description | Default |
|---|---|---|---|
| Video URL | string | Video URL to moderate. Supports MP4, MOV, and WebM formats. | https://d3adwkbyhxyrtq.cloudfront.net/ai-videos/sample.mp4 |
| Context Text | string | Optional context or custom moderation criteria to bias the moderator. | |
Video URL to moderate. Supports MP4, MOV, and WebM formats.
https://d3adwkbyhxyrtq.cloudfront.net/ai-videos/sample.mp4Optional context or custom moderation criteria to bias the moderator.
Developer documentation
Submit the video: POST video_url (MP4, MOV, or WebM) and an optional text describing custom moderation criteria.
Wait for the verdict: Poll /api/v1/predictions/{request_id}/result or pass a webhook query param. Video inference scales with clip length — expect a handful of seconds for short clips.
Read the result: output.flagged is the top-level gate. output.categories enumerates which categories were tripped (e.g. sexual, violence), and output.outputs carries the raw detection strings from the moderator.
Frequently asked
MP4, MOV, and WebM hosted at a public URL.
There's no hard length cap, but inference time grows with clip length. Trim previews to the most representative segments for the fastest results.
Harassment, hate, sexual, sexual involving minors, and violence — plus any custom criteria you pass via `text`.