Live deepfake detection index
Real results from media the public runs through our free detectors — how much is flagged AI, which generators we actually catch, and where suspect clips come from. Every number below is measured, not modeled. It's early and small; it grows as more people check.
Read this before quoting a percentage. This is a small, self-selected sample: people run media through the tool because they already suspect it. So the 36% flag rate (186 of 518 checks) reflects that suspicion, not how much of the internet is fake. Treat these as transparent detector logs, not a population estimate.
Notice on Probabilistic Forensics: All counts reflect automated model confidence scores crossing detection thresholds. They describe detected algorithmic signals and do not constitute definitive factual certifications or legal determinations.
Generators we've actually identified
measured, not ranked by market shareThe generative models our detector attributed on flagged media so far. Counts are deliberately shown raw — with a sample this size, a share percentage would be noise.
| Model identified | Media type | Times detected |
|---|---|---|
| ByteDance Seedance 2 | video | 33 |
| Gemini 3 | image | 16 |
| GPT-Image 1.5 | image | 11 |
| Unattributed | image | 11 |
| Bytedance Seedance 2 | video | 10 |
| Stable Diffusion | image | 10 |
| Alibaba Wan | video | 7 |
| Ai Audio | video | 7 |
| Qwen | image | 7 |
| Openai Gpt Image | image | 6 |
| MiniMax Hailuo | video | 5 |
| Unattributed | video | 5 |
| Flux (Black Forest Labs) | image | 5 |
| C2pa Synthetic | image | 5 |
| Flux | image | 4 |
| Sora2 | video | 3 |
| Alibaba Wan | image | 3 |
| 4o | image | 3 |
| Veo3 | video | 2 |
| Minimax Hailuo | video | 2 |
| Higgsfield | image | 2 |
| Flux2 | image | 2 |
| Meta Movie Gen | image | 2 |
| xAI Grok Image | image | 2 |
| Meta Movie Gen | video | 1 |
| Kling Ai | video | 1 |
| Gemini 3 | video | 1 |
| Imagen4 | video | 1 |
| Ltx | video | 1 |
| Krea | video | 1 |
| Tencent Hunyuan | image | 1 |
| Bytedance Seedance 2 | image | 1 |
| Minimax Hailuo | image | 1 |
| Mai | image | 1 |
| Kling AI | image | 1 |
Where submitted video came from
How people brought video to the detector — a direct upload, or a link to a platform we resolve.
Methodology & limits
- Every figure is a live aggregate over real checks run through our public detectors since September 10, 2026, recomputed daily.
- Verdicts come from multi-model frame and spectral analysis; a file is “flagged” when AI-generation, face-swap, or synthetic-voice signals cross the detection threshold.
- Our own testing traffic is excluded. The sample is self-selected — users check media they already doubt — so figures describe what people bring us, not the internet at large.
- The sample is currently small (518 files). We publish raw counts and show the sample size everywhere precisely so nobody over-reads a percentage. It gets more meaningful as volume grows.
Cite This Data
Free to cite — please keep the sample-size context. Every figure is measured, not estimated.
Sealed Rose. (2026, October 29). Live Deepfake Detection Index [Data set]. https://sealedrose.com/trackerHave media you need to verify?
Check videos, voice notes, or face images with Sealed Rose's free multi-model forensic pipeline — and your check becomes part of this index.