That little sparkle icon in the corner was originally there to tell everyone "this was made by AI." Now Google has decided that can be optional.
Google VP Josh Woodward announced that users will be able to remove visible watermarks from AI-generated images and videos with a simple toggle. The feature will roll out first on three models—Nano Banana, Omni, and Lyria—before eventually reaching Gemini and the video editing tool Flow. All these models currently mark AI-generated content with the same sparkle icon—and now, that icon is something you can switch off.
The original reason for visible watermarks was straightforward: let anyone instantly recognize that an image or video was AI-generated. Once removal is allowed, that "instant recognition" function is largely dismantled. Woodward's explanation is that "we're balancing creative freedom with safety," implying the toggle is mainly built for creators who need clean output.
That said, the mechanisms actually used to trace the origin of AI content haven't been touched—the invisible SynthID watermark and C2PA metadata still have no opt-out option. The difference is that both of these methods require extra steps to verify, like feeding the file into a specific tool or uploading it to a verification page, unlike the sparkle icon that could be spotted with the naked eye.
The toggle won't be available worldwide. Woodward specifically noted that in any region where the law requires visible labeling, the removal option won't be offered. The EU is one such region—it recently passed rules requiring visible watermarks on "highly realistic" deepfake content, and South Korea rolled out similar regulations earlier this year.
Meanwhile, the AI industry remains far from unified on how to label AI-generated content. Suno just launched a new "transparency tool" that keeps labels visible even when AI-generated songs are shared on other platforms; Anthropic, in order to comply with EU regulations, has started embedding invisible C2PA-based watermarks into text and files generated by Claude. Google's latest adjustment essentially splits "visible labeling" and "traceable evidence" into two completely separate paths.
