Photoshop writes C2PA Content Credentials whenever you use a generative AI feature — Generative Fill, Generative Expand, or similar — even on an otherwise untouched real photograph. That means a genuine photo with one small AI-assisted edit (removing a background object, extending a crop) can carry the same provenance signal as a fully AI-generated image, which is a major reason photographers have reported real photos getting flagged.
Firefly (and Firefly-powered features inside Photoshop) record a digitalSourceType reflecting how much AI was involved. A photo that's 95% unedited but had one object removed with Generative Fill can get a manifest reflecting that AI was used in an edit — not that AI created the whole image — but downstream platforms often don't preserve that nuance when they render a label.
created/converted steps.If you're a photographer or editor who used a single AI-assisted retouch on an otherwise authentic photo, and you don't want the finished file to carry a manifest that could get it auto-labeled as "AI info" on social platforms, you need to remove the Content Credentials before export — Photoshop doesn't offer an easy built-in toggle to suppress this metadata.
AiEraser's View mode shows exactly what's embedded — including any C2PA actions and the XMP edit history — before you decide, and Remove mode strips all of it via in-browser canvas re-encoding.
Sources: public reporting on Meta's AI-labeling policy noting Generative Fill edits can trigger "AI info," and Adobe's public Content Credentials documentation. Adobe's exact controls and defaults change between versions — verify current behavior in your installed version.