Google announced on August 14 that users can now toggle off the visible "sparkle" watermark on AI-generated images, videos, and songs created in Gemini. The change, confirmed by Josh Woodward, Google's VP for Gemini, applies to the company's Nano Banana image model, Omni video model, and Lyria music model — and keeps the invisible SynthID watermark and C2PA content credentials intact.
The new control lives in Gemini's Settings menu under a "Media Watermark" section. Flipping it off removes the small sparkle icon that previously appeared in the bottom-right corner of every AI-generated output. The setting applies to newly generated content only and does not retroactively remove watermarks from previously created files.
What Changes and What Doesn't
For years, Google required a visible watermark on all Gemini-generated media — even for paid subscribers. Users who wanted clean images had to resort to third-party tools or manual editing. That changes now, but only at the surface level.
Google emphasized that the toggle affects only the visible watermark. Every file still carries SynthID, DeepMind's imperceptible watermarking system, plus C2PA metadata that records the content's AI origin. These signals allow detection tools — including Gemini itself and Google Search — to verify whether a piece of media was AI-generated, regardless of the toggle setting.
The feature is rolling out over the coming days in the standalone Gemini app and Google Flow, with Search support promised soon. Google may restrict the toggle in jurisdictions where local law requires visible AI labeling on generated content.
What This Means for Creative and Companion AI Users
The policy reversal is significant for anyone using Google's generative tools for content creation, including companion app developers who integrate image or video generation. The visible watermark was a friction point for creators who wanted clean, professional-looking outputs without a telltale AI badge in the corner.
For NSFW AI and adult content creators specifically, the change removes one barrier to using Google's models — though the company's content safety filters remain in place, and SynthID still enables downstream detection. The invisible watermark means platforms can still identify AI-generated media even when the visible mark is gone, which has implications for content moderation and provenance tracking.
The move also signals a broader industry shift. As AI-generated media becomes ubiquitous, platform companies are wrestling with how to balance transparency labeling with user demand for clean outputs. Google's approach — making the visible mark optional while keeping invisible detection signals — could become a template for other platforms.
Broader Context
The change comes as the EU AI Act's transparency requirements take effect, mandating that AI-generated content be detectable. Google's decision to keep SynthID and C2PA metadata mandatory while making the visible watermark optional suggests the company is betting on technical detection signals over conspicuous labeling.
It also follows years of user complaints. Google support threads dating back to 2025 show users repeatedly asking for a way to remove the watermark, with Google consistently responding that no such setting existed — even for paying subscribers. The August 14 update finally addresses that demand, though only for future generations.