About Image Evidence
Image Evidence helps people check whether an image or video was made or changed with AI. We work from the evidence a file carries and the context around it, and we say plainly where that evidence stops.
What we offer
The free checker reads a file inside your browser and reports what it says about itself: C2PA Content Credentials and their validation state, AI-generation declarations in metadata, and Stable Diffusion style parameters in PNG files. It does not analyze pixels or give an AI probability, and it needs no account.
Alongside the checker we publish guides: one for each major AI image and video generator, four on social platforms, and general guides to provenance tools, metadata, AI-edited photos and AI art. The optional Pro plan adds a third-party detector score from Sightengine, batch checks and saved reports. A developer API is at the waitlist stage.
How we write and review guidance
Every guide is built from primary sources: the documentation and help pages of the companies that make the generators and run the platforms, published standards such as C2PA, and peer-reviewed or preprint research. Guides link their main sources beneath the paragraphs those sources support, so you can check them yourself.
When we cannot confirm that a generator documents a signal, the page says the coverage is unverified rather than assuming the signal is present or absent. We describe visual clues and detector scores as prompts for further questions, not as proof, because both fail on real photographs and on new generators.
Each guide shows the date its guidance was last reviewed. We review a page again when a company changes the documentation it relies on, and when a reader reports a problem. Generator and platform names are used descriptively: Image Evidence is independent and has no affiliation with any of them.
Corrections
If something on the site is wrong or out of date, email hello@imageevidence.com with the page and, if you can, the source that shows the problem. We check the report against the source, correct the page when it is wrong, and update its review date.
How we handle accuracy
We do not publish an accuracy number for the free checker, because it does not estimate whether an image is AI-generated; it reports provenance records that are either present or not. We tested an on-device model of our own, and it is not enabled: on a licensed test set it flagged too many real images as AI-generated, especially older phone photos and screenshots.
Pro detector scores come from Sightengine, and we have not yet published our own evaluation of them. The accuracy page explains the evaluation method any model must pass before we would offer our own estimate.
Your files
The checker reads your image inside the browser tab. The file, its name and its metadata are not uploaded to our servers or written to browser storage. A Pro detector score is the only feature that uploads an image, only when a signed-in user asks for one, and the image is not stored. We do not train on user content. The privacy notice has the details.
What we will not build
We do not offer watermark removal, detector evasion, Humanizer features, face recognition, image-to-person search, or pages that accuse named real people of faking images. A checker result or detector score should never be the only basis for a decision about a person; our terms of use set out how results may be used.
Contact
For questions, corrections, privacy requests or press, email hello@imageevidence.com. Payments for Pro are processed by Paddle.com and Waffo Pancake as merchant of record.