Score every image upload for synthetic manipulation before it reaches your users. Where the file still carries signed content credentials we verify them cryptographically; where it does not, the detection model scores the pixels.
1,000 free credits · no credit card · PNG, JPG, GIF, WebP up to 4MB (10MB by URL)
/api/v1/detect/imagecurl -X POST https://www.wasitaigenerated.com/api/v1/detect/image \
-H "Authorization: Bearer YOUR_API_KEY" \
-F "file=@/path/to/profile-photo.jpg"Content-Type — multipart/form-data
file — required, PNG · JPG · JPEG · GIF · WebP
Cost — 300 credits per image
{
"isAI": true,
"confidence": 0.96,
"verified": false,
"c2pa": null,
"patterns": [
"high_ai_confidence",
"synthetic_artifacts"
],
"analysis": {
"likelihood": "high",
"reasoning": "AI-generated imagery detected with
96.0% confidence."
},
"detailed": {
"scores": { "ai": 96.0, "human": 4.0 },
"model": "v1.1"
}
}The public API covers both. POST /api/v1/detect/image checks stills, including C2PA content credentials. POST /api/v1/detect/video scores a whole clip across sampled frames, or a single frame such as a thumbnail or screen grab.
Upload clips up to 4MB directly, or pass larger ones (up to 50MB) as a URL. Fake profile pictures, generated ID photos and synthetic claim evidence usually arrive as stills; viral clips and impersonation videos go through the video endpoint. See the video docs.
Most detectors hand you a probability and leave you to defend it. This one tells you when it has actual proof.
Read the evidence behind an image
Workflow illustrationModel analysis
An estimate from the image itself. Read the confidence alongside the AI/human assessment.
Content credentials
When present, inspect the C2PA signature and declared source. Check what the credential actually establishes.
Generators following the C2PA standard sign their output with content credentials. When the signature is intact we validate it — the image is AI-generated, and that is not a judgement call.
Use it when the result has to hold up in front of a customer, a moderator or a regulator.
Most images in the wild have been screenshotted, re-encoded or stripped of metadata — which is exactly what someone hiding a deepfake does. For those, the model scores the pixels.
Strong evidence — set a threshold and keep a human in the loop at the boundary.
Multipart upload, Bearer auth, JSON back
curl -X POST https://www.wasitaigenerated.com/api/v1/detect/image \
-H "Authorization: Bearer YOUR_API_KEY" \
-F "file=@/path/to/profile-photo.jpg"isAI — boolean verdict.confidence — 0–1 from the model.verified — signed credentials prove AI generation.detailed.scores — AI vs human as percentages.400 — no file, bad format, over 4MB.401 — bad or missing key.402 — not enough credits (insufficient_credits) — fewer than 300 left.403 — website-only plan (Pro); the API runs on credits or API Business.429 — over 120 requests/min per key; honor Retry-After.Generated profile photos are the cheapest input to a romance scam. Screen them at signup, where a fake account costs nothing to stop. See the use case
Flag synthetic ID photos and document images during onboarding, alongside your existing liveness checks. See the use case
Check damage photos at intake, before an adjuster builds a payout on generated evidence. See the use case
Score every upload as it lands and route synthetic media into the moderation queue you already run.
Verify user-submitted and wire imagery before publication, and keep the result as part of your editorial audit trail.
Check imagery attached to executive impersonation and invoice-fraud attempts before someone acts on it.
300 credits per image — no subscription needed to start.
1,000
credits on email verification
3 images to try the endpoint. No card required.
from $10
one-time, never expires
50,000 credits ≈ 166 images. 1,500,000 credits ≈ 5,000 images.
$299/mo
3M API & MCP credits every month
Everything in Pro plus the API and MCP server, white-label PDF reports and priority support. Cancel anytime.
Yes. POST /api/v1/detect/video takes a clip (MP4, MOV, WebM, AVI) and scores it across sampled frames, or a single still such as a thumbnail or screen grab. Upload files up to 4MB directly, or pass larger clips up to 50MB as a URL. A clip is charged 300 credits per analyzed frame (16 to 50 frames, so 4,800 to 15,000 credits); a single frame costs 300.
Any image that is synthetic or synthetically manipulated — a fully generated face, a swapped face, a generated scene presented as a photograph. The endpoint returns a probability that the image is machine-generated rather than a taxonomy of the manipulation technique.
confidence is the model’s probability that the image is synthetic. verified means the file carries a C2PA content credential whose signature verifies and which declares the asset AI-generated, which proves it. The parsed credential is returned as c2pa (vendor, generator, issuer, signedAt, signatureValid, sourceType, isAI) or null when there is none — a valid credential with isAI: false and sourceType digitalCapture is a signed camera photo, useful evidence in the other direction. verified: true is certainty; confidence is strong evidence that still deserves a human decision at the boundary.
Because screenshotting strips the content credentials. The signature lives in the file’s metadata, and a screenshot is a new file, so c2pa comes back null. You still get a model-based confidence score, but the cryptographic proof is gone. Always prefer the original upload over a re-captured copy.
Yes, and that is the usual pattern: call the endpoint at upload time, store isAI, confidence, verified and c2pa.vendor alongside the asset, and route anything above your threshold into your existing moderation queue. Call it asynchronously — do not block the user’s upload request on it.
300 credits per image. Verified accounts start with 1,000 free credits, which covers three test images. The $10 pack of 50,000 credits is roughly 166 images, the $49 pack of 300,000 is 1,000, and the $199 pack of 1,500,000 is 5,000. For steady volume, API Business is $299/month with 3,000,000 credits every month (≈ 3M words or 10,000 images), white-label PDF reports and priority support; overage draws from credit packs.
PNG, JPG, JPEG, GIF and WebP, up to 4MB (10MB by URL), sent as multipart/form-data with the field name file. Downscale very large images before uploading for the most reliable throughput.
Less reliable than on originals, and worth designing around. Platform re-encoding destroys fine detail the model relies on, so scores on a heavily compressed repost are weaker evidence than scores on a source file. Set your threshold with that in mind and keep a review step for consequential decisions.
Create an account, verify your email, and your API key is waiting in the dashboard.