Was it made by AI?
Drop a picture, a recording or a video. This page reads what the file says about itself: whether its metadata or its Content Credentials say a model made it, and whether it carries Peak’s invisible watermark — and on which day the file was made. The file is read in your browser and never leaves your device.
What it reads
- Metadata. Peak writes into every picture, recording and video it generates that a model made it, which model, and when — the IPTC digital source type that photo agencies and C2PA use. Other apps write the same.
- Content Credentials (C2PA). Signed provenance some model providers attach, such as OpenAI and Anthropic. This page sees whether they are there and whether they say a model made the file; it does not verify the signature — AI Check in Peak does. Peak signs what it generates from 1.8.1.
- Peak’s invisible watermark. Where the model that made something does not watermark its own output, Peak adds a mark you cannot see or hear — in the picture, in speech, and in every frame of a video — saying it came from Peak and the day it was made. It survives compression, resizing and re-encoding. This page reads the mark across the whole picture; AI Check in Peak also finds it in a cropped one.
What it cannot tell you
Finding nothing does not prove a person made a file. Metadata disappears when a file is saved again elsewhere, a cropped picture needs AI Check in Peak to find its watermark, and other companies’ watermarks — such as Google’s SynthID — need their own detectors. Treat a marking as evidence that something was generated; never treat its absence as evidence that it was not.
Why this page exists
Under the EU AI Act, content made with generative AI has to be marked so that software can recognise it — and anyone should be able to check, not only people who have the app that made it. Peak marks what it makes; this page lets everyone read the mark. It is also inside Peak, under Intermission → Tools → AI Check, and you can ask Peak AI “is this picture AI-generated?”