Social media
Social networks, social-listening and brand-safety tools, and agencies all deal with images and clips that may have been generated. Our API scores them for AI generation from the pixels, including copies that were screenshotted and re-uploaded many times over.
99.5%
of AI images caught with only 1% of real photos flagged, across 19 current generators
96.2%
still caught after WhatsApp, Instagram and re-share compression
0.987
AUROC on AI video from 12 video models
Hidden split of our October 2026 benchmark, never used in training. Method and full results
A viral image rarely arrives as the original file. It has been screenshotted, cropped, re-uploaded and recompressed by each platform along the way, and any metadata is long gone.
Generated profile photos give fake accounts a believable face. Networks of them push AI-made posts and ads, or impersonate brands and public figures.
A fabricated image of a public figure or a staged clip can travel widely within hours. During election periods, the cost of reacting late is especially high.
A probability that an image was generated or synthetically altered. The model is trained on images that went through WhatsApp, Instagram and re-share compression, so recompressed copies still get a usable score.
Score the avatar of a new or suspicious account as part of your fake-account signals, alongside the behavioural data you already have.
Send a clip by URL, up to 50MB, or upload one up to 4MB. Frames are sampled across the clip and the response reports how many were analysed and how many were flagged.
Where a post still carries a C2PA manifest, it is parsed and its signature verified, so a signed AI declaration from the generator can be shown to users as a label.
Send images and clips from your ingestion, listening or ad-review pipeline to the REST API as files or URLs.
You choose the cut-offs. A brand-safety report can use a different threshold than an ad-approval queue or a fake-account model.
Label posts, escalate likely impersonation to a reviewer, and include scores in client or brand-safety reports with the frame counts for video.
Score a viral clip by URL
curl -X POST https://www.wasitaigenerated.com/api/v1/detect/video \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"url": "https://example.com/media/viral-clip.mp4"}'Images, single frames and whole clips. See the API docs for responses, limits and pricing.
Yes. Recompressed copies are part of the training data. On our published benchmark the model still caught 96.2% of AI images after WhatsApp, Instagram and re-share compression, at a 1% false-positive rate.
A detection score is a probability, not proof. Use it as one signal among others in your fake-account review, and have a person confirm before you act against an account or call something impersonation.
On the same benchmark images, our model caught 99.5% of AI images at a 1% false-positive rate across 19 current generators. ZeroGPT, which uses TruthScan for image detection, reached 97.2%. The method and full results are published.
Across 12 video models the video detector reached an AUROC of 0.987. On the benchmark clips it caught 60 of 63 AI clips and flagged 1 of 60 real clips.
We run paid pilots on a sample of your images and clips, calibrate the threshold to your tolerance for false positives, and report detection and error rates back before you commit. Volume pricing and an SLA are available for production, and the models can run on your own GPUs and servers.
Talk to us about a pilot