Home/Guides/Deepfake Detection
Tool
Find out if an image was made by AI, or if the face was swapped.
Deepfake Detection checks any image - a selfie, an ID photo, a damage photo, a screenshot of a document - for signs that it was generated by an AI model or that a face was swapped in. You get an authentic, AI-generated, face-swap or uncertain verdict, a risk score, and the specific findings behind it.
- No document storage
- Digital, scanned or photographed
- ~1 minute per verdict
Last updated:
The problem
Why a visual check no longer catches synthetic images
Current image models produce photos with correct hands, plausible lighting and legible text. The tell-tale artefacts reviewers were trained to spot are mostly gone, and a face swap on a real photo leaves the rest of the scene genuine.
AI-generated images were never captured by a camera. They lack the sensor noise, colour-filter pattern and lens behaviour every real photo carries, and many still carry provenance the generator wrote into the file.
Face swaps start from a real photo, so whole-image checks can pass. The evidence sits at the blend boundary: skin texture, sharpness and lighting that change where the new face meets the original frame.
Deepfake Detection is a separate strategy from Tamper Check. Tamper Check looks for edits to a genuine document; Deepfake Detection asks whether the image itself is synthetic.
What we detect
Forensic signals that expose fraud
Content credentials (C2PA)
Reads signed C2PA manifests and validates the chain. A valid credential naming an AI tool is proof of generation, not a guess.
Invisible AI watermarks
Checks for the hidden pixel-level watermarks some generators embed, which survive screenshots and re-saves that strip metadata.
Generator fingerprints
Matches output dimensions, generation-parameter chunks, software tags and IPTC source types left by Midjourney, Stable Diffusion, DALL-E, Firefly and others.
Camera-sensor forensics
Measures sensor noise (PRNU), colour-filter-array patterns and chromatic aberration. Real cameras leave these; generators do not reproduce them faithfully.
Face-swap blend analysis
Inspects the face boundary for texture, sharpness and compression breaks, and compares face lighting with the shadows in the rest of the scene.
Vision adjudication
A vision model reviews the image alongside every forensic measurement - anatomy, text, reflections, physics - and explains its verdict in plain language.
The solution
One upload, a clear verdict on whether the image is real
Run Deepfake Detection from the dashboard for a single image, or set strategy to deepfake_detection on the API to check every selfie, ID photo or claim photo in your workflow.
How it works
- Four verdicts - authentic, AI-generated, face swap or uncertain
- Images of any kind - JPG, PNG, WebP, HEIC, screenshots and phone photos
- Provenance first - signed credentials and watermarks settle the verdict when present
- Findings with regions - see where the evidence is, not just a score
- Same API and dashboard as Tamper Check - one parameter switches strategy
- Per-image pricing - no subscription
Why us
What you won't get elsewhere
Findings, not a score
We name the field that changed and where it sits on the page.
Others hand you a number to justify.
One endpoint, every type
100+ document types, 190+ countries, no per-type setup.
Elsewhere, each new form is a project.
Catches what was never real
A dedicated pass for AI-generated and deepfake documents.
Template matching only catches edits.
Nothing is retained
Analysed in memory, then discarded. Never trained on.
Most services keep documents for QA.
Live in minutes
Published pricing, no minimum. Test it before you talk to us.
Platforms want a contract first.
Sits alongside your stack
Keep Onfido or Jumio for face match. We add the forensics.
Not a rip-and-replace.
FAQ
Common questions
What does Deepfake Detection check?
Whether an image was generated by an AI model or had a face swapped in. It reads content credentials and invisible watermarks, matches known generator fingerprints, measures camera-sensor noise and lens behaviour, analyses the face boundary for blending, and has a vision model weigh all of it.
How is this different from Tamper Check?
Tamper Check looks for edits to a genuine document - a changed balance, a re-typed date, a pasted signature. Deepfake Detection asks whether the image was ever captured by a camera at all, or whether the face in it is someone else's. Use Tamper Check for PDFs and documents, Deepfake Detection for photos and selfies.
Does it work on PDFs?
No. Deepfake Detection runs on images. For PDFs, Tamper Check already includes AI-generation signals for documents that were fabricated rather than edited.
Can it tell which AI tool made the image?
Often. When the image carries content credentials, a generator watermark, or generation metadata, the finding names the tool. When those have been stripped, the verdict rests on pixel-level evidence and the tool may not be identifiable.
How do I use it through the API?
Upload the image to the same documents endpoint you use for Tamper Check and set strategy to deepfake_detection. The response carries the verdict, risk score, confidence, findings and a plain-language summary.
Or skip the call entirely
We'll match your first top-up in credits, so you can run your own documents through it now.