Как на самом деле работает происхождение изображений, что делают стандарты и где каждый метод обнаружения упирается в предел.
SynthID Survives a Screenshot, But Only If You Crop It First
We screenshotted a Gemini image and asked Gemini about it twice. Framed, it said no watermark. Cropped tight, it found one. Same screenshot, opposite answers.
EU AI Act Article 50: Labelling AI Images
EU AI Act Article 50 applies from 2 August 2026: synthetic images must carry machine-readable marks. What that means in practice, and what it does not.
How to Check a Photo from a Listing or a Dating Profile
Generated pictures cost nothing to make, which makes them the cheapest part of a scam. Here is what you can actually establish about a photo somebody sent you, in the order worth trying.
How to Check If an Image Is AI-Generated
A practical guide to checking an image for AI origin: read its C2PA, IPTC and generator metadata, verify signatures, and know what absence proves.
Read Stable Diffusion Metadata from a PNG
Stable Diffusion writes prompt, seed, sampler and model name into PNG text chunks. Here is how those chunks work and four ways to read them back out.
SynthID Explained, and Its Limits
SynthID hides a watermark in the pixels of Google AI images. Google's own app can now read it back. Here is how it works, how to check, and what it cannot do.
Verifying an Image in a Newsroom, Without Uploading It
A working order of checks for images from sources: provenance first, then reverse search, then the file itself. Plus how to describe what you found without overstating it.
What Are C2PA Content Credentials?
C2PA Content Credentials explained: what a signed provenance manifest contains, how it survives editing, when it breaks, and how to verify one yourself.
Why Instagram Strips Your Image Metadata
Instagram, WhatsApp and most social platforms re-encode every upload and drop EXIF, XMP and C2PA data. Here is what gets destroyed and what survives.