How screenshot forgery detection differs from photo detection
A screenshot is not a photograph — it's a flat digital capture of a screen. That distinction matters more than it sounds: heuristics built for AI-generated photos (like checking for camera sensor grain) are meaningless here, because a genuine screenshot never has sensor noise to begin with. This tool uses a different set of checks, built specifically for how screenshots are made and edited.
File format
Genuine screenshots are almost always saved as PNG. A JPEG claiming to be an untouched screenshot is a flag on its own.
Screen resolution match
Real screenshots match a real device's exact screen resolution — generators and edited crops often don't.
Editing-software trace
Files that passed through Photoshop, GIMP, Canva or similar tools can leave a fingerprint in the metadata.
Localized ELA hotspot
Recompression analysis run block by block — a locally edited region (changed amount, altered name) stands out from the rest of an otherwise uniform image.
Why "localized," not "global"
A generic photo detector looks at the average compression error across the whole image. That approach fails on screenshots: flat UI graphics naturally compress very uniformly, so a low overall error is normal, not suspicious. What actually gives away an edit is a hotspot — one small region (say, a transaction amount or a name) with a compression signature that doesn't match the rest of the image, because it was pasted in or retouched after the original capture.