Bureau's Deepfake Detection uses an ensemble of 4 specialized models — GAN fingerprint analysis, face-swap detection, frequency-domain analysis, and a Bureau ensemble model — to catch manipulated and AI-generated images in under 500ms.

Three outputs per request: a binary verdict (authentic / deepfake), a confidence score (0-100), an explanation of what was found, and a manipulation heatmap showing exactly which regions of the image were flagged.

Covers face swaps (StyleGAN, DeepFaceLab, FaceSwap), ID document tampering (photo splicing, font manipulation), and fully AI-generated images from Midjourney, DALL-E, SDXL, and Flux. Bundled into Bureau's Identity Verification and Merchant Onboarding workflows.

▶ Toggle Genuine / Fraud in the header, then click Analyze

See the full ensemble analysis and manipulation heatmap for each mode.