Sightengine
AI-made and face-swap signals.
Scores images and short videos for signs of AI generation and face swaps, and recordings for AI-made speech. These scores decide the authenticity label together with Content Credentials.
01What it does
- Images: two scores, AI generation (
genai) and face swap (deepfake). - Videos up to 60 seconds: the same two scores for sampled frames, through Sightengine's synchronous video endpoint. The app shows how many sampled frames scored above your threshold and the highest score, never frame times.
- Audio, voice messages and the sound of a video: an AI-speech score (
ai_speech). The sound of a video needs ffmpeg on the server.
How the scores become the authenticity label:
| Scores | Label |
|---|---|
| The highest score at or above Likely (default 0.80) | Likely AI-made or altered (Unclear when a trusted Content Credentials signer says otherwise) |
| Every score at or below No signs found (default 0.20), nothing decisive left unchecked, and no broken Content Credentials record | No signs found |
| Anything else, or no score at all | Unclear |
Set both scores in the console under Media caps and thresholds. A score is a probability, never proof, and the app says so. The authenticity axis never reaches Danger: strong signs are shown at Caution.
When Sightengine does not answer, the media units of that check are given back to the person's daily allowance.
02Get the keys
- Create an account on Sightengine and choose a plan (video needs a paid plan).
- Open the dashboard and copy your API user and API secret.
03Switch it on
- Console, AI media signals (Sightengine): turn on Use Sightengine, paste the API user and the API secret (both are stored encrypted and never shown again).
- Press Test connection. It sends one small image made by the server; it counts as one image check on your account.
04Terms
Files are sent to Sightengine for scoring on your own account, so Sightengine is named on the app's consent screen. Read Sightengine's terms and data processing terms for your country before you launch.
05Cost
Billed to your Sightengine account in operations. One image check uses 10 operations (two models of 5 each). A video uses the same per sampled frame. AI-speech detection uses 1 operation per started 10 seconds. As read on 11 November 2026: a no-cost tier of 2,000 operations a month, about 200 image checks, and paid plans from about $29 a month; video checks need a paid plan. Check Sightengine's pricing page before you launch.