FAQ

Questions, answered

How does TruthLens detect AI-generated images?+

We combine a vision foundation model trained to spot generative artifacts with heuristic checks for things like impossible reflections, melted text, asymmetric anatomy, and characteristic diffusion noise.

Which generators can it identify?+

Midjourney v5/v6, Stable Diffusion 1.5/XL/3, Flux, DALL·E 3, Gemini Image, Sora image previews and several open-source variants. New models are added continuously.

Can it spot deepfakes?+

Yes. We specifically score face-swap consistency, identity replacement, and AI inpainting on portraits — common in deepfake content.

Is there a confidence score?+

Every report includes a 0–100% confidence value, a risk level (Low/Medium/High) and a list of concrete artifacts that drove the verdict.

Do you keep my images?+

No. Uploads are processed in-memory only and discarded immediately after analysis. We never use your images for model training.

Do you have an API?+

Yes. The Pro plan includes a REST API with the same JSON shape returned by the playground. Enterprise adds dedicated infra and SLAs.

What file types are supported?+

JPG, PNG and WebP up to 6MB. Larger files and video frames are available on Enterprise.

Can it be wrong?+

Yes — no detector is perfect. Real photos with heavy editing or low-light noise can confuse classifiers. Use the reasoning section to evaluate the verdict yourself.