Google now lets you hide the visible watermark on its AI images and video

A new toggle removes the sparkle mark from Gemini, Flow, and soon Search — but invisible tracking codes stay embedded no matter what.

What happened: Google announced Friday it's rolling out a setting that lets people switch off the visible "sparkle" watermark that normally sits in the corner of images, video, and music made with its Nano Banana, Omni, and Lyria AI models. The toggle is live now in Gemini and the Flow video editor, under Settings > Media Watermark, with Search support coming soon. It doesn't touch the invisible SynthID watermark or C2PA metadata baked into every file. Google is also open-sourcing a checker tool called Credentio so outside developers can build their own verification into apps.

Why it matters: Visible watermarks were meant to flag "this is AI-made," but they also make content look unfinished for real creative or professional use, and plenty of people were already cropping or editing them out anyway. Google's move formalizes that reality instead of fighting it, following OpenAI and Meta, neither of which ever used visible marks, leaning instead on invisible signals. The tradeoff is that spotting AI content now takes an extra step, actually asking Gemini or Search, instead of a glance.

How it works, plainly: SynthID hides a statistical pattern inside an image or audio file's pixels or waveform; C2PA attaches metadata describing how and when it was made. Both survive even when the visible sparkle is switched off, and tools like Gemini or Search can read them to confirm AI origin. Credentio lets other apps run that same check locally rather than depending on Google's own servers. Countries that legally mandate visible labels won't get the option to turn it off.

The rollout: The change is rolling out over the coming days for Gemini and Flow, with Search to follow, though it's blocked in countries with visible-labeling laws. It lands the same week Anthropic said it would add invisible watermarks to Claude's text and files to meet EU transparency rules, a move that drew its own criticism. Together, the two moves point to AI companies settling on invisible-by-default labeling while treating visible marks as optional, not mandatory.