AI video detector
Sora video detector
OpenAI signs Sora videos with Content Credentials and adds a visible watermark in the Sora app. Drop a clip to read the credentials and analyse its frames — or to find out whether the labels have been stripped.
Free · no sign-up · your file never leaves your device
Drop the suspected Sora clip or click to choose
MP4 · MOV · WebM — 8 frames are sampled and checked on your device
Sora’s labels, layer by layer
OpenAI released Sora to the public in late 2024 and launched a dedicated Sora app in 2025, built around a social feed of generated clips. Because the app made realistic video of real people — including users’ own likenesses — trivially easy, provenance mattered from the start. OpenAI uses three layers:
- A visible watermark that moves around the frame in downloaded videos, so it can’t simply be cropped from one corner.
- C2PA Content Credentials embedded in the MP4, signed by OpenAI and recording that the video was generated by AI.
- Internal detection that OpenAI says it runs to recognise its own output, which isn’t available to the public.
GPTTrace reads the second layer directly. If the manifest is present and signed by OpenAI, the result is marked as provenance. The frame strip helps you look for the first.
When the labels are gone
A clip that has been screen-recorded or re-uploaded loses its credentials, and third-party tools that blur or inpaint watermarks are easy to find. Then the evidence is in the footage:
- Duration and structure. Sora clips are typically short, and longer videos are often several clips joined with hard cuts.
- A smudged patch moving through the frame where a watermark was removed.
- Continuity errors — objects appearing, text changing, extra limbs in crowds, physics that almost works.
- The account. Feeds that post only spectacular, short, perfectly framed clips with no behind-the-scenes material are often AI channels.
GPTTrace’s frame classifier adds a statistical check on top. Use the original download whenever you can find it; the further a copy is from the source, the less any detector can see.
Sora and real people
The Sora app’s cameo feature means many viral clips show real, identifiable people — sometimes the creators themselves, sometimes public figures who opted in, sometimes people who didn’t. A convincing clip of a celebrity or politician saying or doing something surprising should be checked against their official channels before you believe or share it. OpenAI applies restrictions and takedown processes for misuse of likenesses, and reporting a misleading clip on the platform where you found it helps.
For businesses, the same technology produces fake product demonstrations, fake customer testimonials and fake “behind the scenes” footage. If a video is the main evidence for a claim you are being asked to pay for, ask for something that is hard to generate: a live call, an independent review, or the original raw footage with its camera metadata.
What Sora and OpenAI leave in a file
These are the traces GPTTrace checks for. “Some files only” means the trace is often missing — a re-save, screenshot or social-media upload removes metadata — so its absence proves nothing.
| Generator | Where | What to look for | How reliable |
|---|---|---|---|
| Sora | C2PA manifest | Content Credentials signed by OpenAI on downloaded videos | Documented by the vendor |
| Sora | Frames | A visible, moving Sora watermark on videos from the Sora app | Documented by the vendor |
| OpenAI (DALL·E / ChatGPT images / Sora) | C2PA manifest | Content Credentials signed by OpenAI, with a c2pa.created action and digitalSourceType trainedAlgorithmicMedia | Documented by the vendor |
| OpenAI (DALL·E / ChatGPT images / Sora) | Pixels | Square or 3:2 sizes such as 1024×1024, 1536×1024 and 1024×1536 | Commonly seen |