How AI Image Detection Works
Every scan runs through multiple independent forensic checks — here's what each one looks for.
AI-generation scoring
Every image, video, or voice recording you upload is scored by our detection engine for AI-generation likelihood. You get a confidence-scored verdict — how likely the file is to be AI-generated or manipulated — rather than a flat yes/no.
Deepfake risk scoring
For images and video containing faces, we return a separate deepfake risk score alongside the AI-generation verdict, so face-manipulation risk shows up as its own signal instead of being folded into one overall number.
Provenance & metadata
When a file carries embedded metadata or content credentials — camera model, editing software, capture date — we read and display it next to our own analysis. It tells you what the file itself claims, which you should weigh alongside the detection verdict rather than treat as independent proof.
Spectral & frequency analysis
For images, we run a real frequency-domain (FFT) analysis of the pixel data, independent of our detection engine. Natural photos tend toward a smooth falloff in frequency energy; the upsampling stages inside GAN and diffusion decoders often leave periodic peaks and extra high-frequency energy behind. We measure both and combine them into a single anomaly score — a supporting signal, not proof on its own.
Model attribution
When a file is flagged as AI-generated with high confidence, we also surface our best guess at which generator was likely used — for example Midjourney or Stable Diffusion — ranked by confidence. This attribution is a secondary signal and is most reliable on unedited files.
Exportable forensic reports
Every scan can be shared via a report link or exported as a PDF, so you can cite, archive, or hand off a result as part of an editorial, legal, or compliance workflow.
No detection method is perfect on its own. That's why we combine multiple signals rather than relying on any single check — and why every report shows a confidence level, not just a yes/no answer.
See which plan fits your workflow
From a free tier to higher-volume enterprise plans — pick what matches how you'll use verification.