Meta stops publicly tagging lightly edited photos as AI

If you used an AI eraser to clean up a tourist in the background of your vacation photo, Instagram used to treat you like you just cooked up a deepfake from scratch.

After months of complaints from photographers and digital artists, Meta is finally dialing it back. On September 12, the company announced an update to how its "AI info" tags work across Facebook, Instagram, and Threads, with the rollout beginning next week.

Moving touch-ups out of the spotlight

Under the new setup, Meta is drawing a hard line between media created by AI and media simply touched up by it.

If Meta's systems detect that a post was genuinely generated by an AI tool—or if the creator voluntarily disclosed it—the visible "AI info" label will stay pinned right on the post. Clicking the badge will tell viewers whether the label came from industry technical signals or the user's own disclosure.

However, if a photo or video was merely edited or retouched using AI-powered tools, Meta is removing that front-facing badge entirely. Instead, the "AI info" note will be tucked away inside the post's three-dot context menu.

This marks Meta's second backtrack this year. Back in April, the platform began aggressively tagging posts with "Made with AI." That triggered backlash in July when professional photographers discovered real photos were being branded as AI creations simply for using modern Adobe editing tools. Meta watered down the text to "AI info" in July, and is now banishing minor edits from public view.

Why it matters

Meta's course correction is a huge relief for creators who were practically being punished for using standard 2024 creative software.

Almost every modern creative app—from Lightroom to basic phone gallery tools—now bakes generative fill or automated denoising into basic workflows. Flagging all of them with a loud warning badge dilutes the meaning of synthetic media warnings and creates unnecessary stigma for real photographers.

At the same time, this highlights how messy content provenance still is across the industry.

Meta relies heavily on shared metadata standards like C2PA, but these signals struggle to distinguish between a harmless skin retouch and a manipulated news image. Moving minor edits into an obscure menu keeps the feed clean, but it also creates a gray zone where altered media can hide behind subtle technicalities. For platforms trying to police election-year deepfakes, finding the line between editing and generating remains an unsolved puzzle.