Video moderation — test

Frames are sampled locally and scored by OpenAI; audio is analysed on this machine. Audio-only files (mp3, wav) are accepted — useful for testing the acoustic branch on its own. Max 40 MB.

What is checked. Frames go to OpenAI; audio is analysed on this machine (speech is transcribed and scored, wordless audio is classified). The result says which halves actually ran — a clean verdict with a NOT CHECKED note means part of it was skipped. CSAM is never detected by this model.

Images

Up to 10 at once. Sent straight to OpenAI — no frame extraction, no audio.

Video or audio

Frames are sampled and scored; audio is analysed on this machine.

Fixed during testing. Change SERVICE_TOKEN in .env to something private before this is reachable from outside.