Google to allow removal of visible watermarks from AI-generated content

Only the visible mark gets switched off

Google is letting users turn off the visible watermark that has been automatically stamped on images, video, and music generated by its models. Josh Woodward, VP of Gemini, said on X that the toggle applies to the Nano Banana (image), Omni, and Lyria (music) models. The setting lives in the Gemini app and Google's video editing tool Flow, with Search support to follow later. The feature is rolling out over the next few days, and can be switched on or off under Settings > Media Watermark.

The invisible SynthID watermark and C2PA standard metadata, however, remain unaffected by this setting. Woodward offered a brief explanation, saying the change "balances creative freedom with safety." On the same day, Google also released an open-source library called Credentio, which lets developers embed their own verification logic directly into their apps.

Why there are two layers of watermarking

SynthID is a watermarking technology built by Google DeepMind that embeds patterns in pixels or audio waveforms that are undetectable to human eyes or ears. This mark largely survives cropping or compression, so running a file through Gemini or Search can confirm whether it was AI-generated. The visible, logo-style watermark that has appeared in the corner of the screen until now is a different matter. It functioned less as an identification tool and more as branding and an immediate visual warning, making it an obstacle for users who wanted to put photos or videos to practical use right away.

Labeling AI-generated content has become an industry-wide talking point recently. Anthropic began adding watermarks to text and files created by Claude in order to comply with EU regulations, a move that drew pushback over concerns it could expose AI use in workplaces or schools. Google's latest move looks like the opposite approach, but in substance it follows a similar logic: leave the visible marking to user convenience, while keeping socially necessary identification functions — such as combating deepfakes or tracing content provenance — running at an invisible layer. C2PA is an international standard that cryptographically attaches a record of capture and editing history to photo and video files, and has been adopted jointly by companies including Adobe and Microsoft.

So what actually changes

Users who want to use a Nano Banana–generated image for ad creative or a portfolio no longer need a separate editing step to remove the logo in the corner. But the fact that the file was AI-generated doesn't disappear. Running a SynthID check or inspecting the C2PA metadata will still reveal the generation history, meaning the surface may look more natural, but traceability remains intact. With Credentio now open-sourced, other developers also have a path to implement similar verification logic directly within their own apps.