Trust & safety
Communities, forums, dating apps and creator platforms receive a steady stream of photos and clips from users. Some of them are generated. Our API gives every upload an AI-generation score, so your moderation rules can act on it and your reviewers can spend their time on the hard cases.
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 fake profile built on a generated face looks like any other member. On dating and community platforms, users stop engaging when they can no longer assume the person behind a photo exists.
Many platforms now require AI content to be labeled or restrict it in certain areas. A rule like that only works if you can tell which uploads are synthetic in the first place.
Every upload is a potential review item. Human moderators cannot look closely at each one, so synthetic media slips through or the backlog keeps growing.
Each image gets a probability that it was generated or synthetically altered. The model is trained on current generators and on files that were compressed and re-shared before they reached you.
Frames are sampled across the whole clip and scored by a dedicated video model. The response tells you how many frames were analysed and how many were flagged.
A cover image or a single frame can be scored on its own with the video model, which helps when you only store a preview.
If an upload carries a C2PA manifest, it is parsed and its signature verified. A valid AI declaration from the signer is reported as certain, which makes it a clean basis for an automatic label.
Call the REST API from your upload pipeline with the file, or with a URL for clips stored in your own bucket.
You choose the thresholds. Label or hold high scores automatically, send the middle band to a moderator, and publish the rest as normal.
Attach an AI label where your policy or the EU AI Act Article 50 transparency rules call for one, and store the score with the moderation decision for appeals and audits.
Score an image from the upload queue
curl -X POST https://www.wasitaigenerated.com/api/v1/detect/image \
-H "Authorization: Bearer YOUR_API_KEY" \
-F "file=@upload.jpg"Images, single frames and whole clips. See the API docs for responses, limits and pricing.
We advise against it. A detection score is a probability, not proof. Use it to label content, prioritise review or hold an upload, and have a moderator confirm before you take action against a person or an account.
That depends on the threshold you set. On our October 2026 benchmark across 19 current image generators, the model caught 99.5% of AI images while flagging 1% of real photos. You can raise the threshold to flag fewer real photos, at the cost of letting more AI images through.
The API is built for automated pipelines. For high, steady volume, enterprise plans include dedicated capacity and an SLA, so moderation latency stays predictable.
Uploaded files are deleted after analysis; the API keeps the result, not the file. A data processing agreement and self-hosting on your own GPUs and servers are available for enterprise customers.
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