Google Lets You Remove Gemini's Visible 'Sparkle' Watermark From AI Images, Video, and Music

Google has added a Media watermark toggle to Gemini and its AI video generator Flow, letting creators remove the visible "sparkle" icon that appears in the bottom-right corner of generated images, videos, and music tracks — a small but practically significant change for anyone delivering clean AI-generated assets to clients.

Key takeaways

• Google's new Media watermark setting in Gemini and Flow lets users toggle off the visible sparkle icon on AI-generated images, video, and audio.
• Disabling the visible watermark does not remove the invisible machine-readable metadata used to identify AI-generated files — that layer persists regardless.
• The toggle applies across Gemini's image generation, the Flow AI video tool, and AI-generated music.
• Creators who need clean, unbranded deliverables — for client mockups, social posts, or editorial use — can now do so without post-processing workarounds.
• Google's approach mirrors a growing industry split: visible branding becomes optional, but invisible provenance signals stay mandatory.

The setting lives inside Gemini's settings menu, and according to The Verge, toggling it off strips the sparkle from all three output types — still images, video clips, and audio. That covers a meaningful slice of a modern AI creator's pipeline in one switch.

What the toggle does — and doesn't — change

The critical detail is what the toggle doesn't touch. As TechCrunch reports, turning off the visible watermark has no effect on the invisible benchmarks embedded in the file — the machine-readable signals that identify the content as AI-generated. Those stay in the file regardless of the setting. Google's move is therefore less about transparency policy and more about surface aesthetics: the badge is optional, the provenance record is not.

This distinction matters a lot in practice. If you're generating a product mockup in Gemini and handing it to a client, the sparkle in the corner has always been awkward — either you crop around it, mask it out in post, or explain AI involvement before the client is ready for that conversation. The new toggle removes that friction without actually stripping the file's identity. The C2PA-style metadata underneath remains, which means AI-detection tools and platforms that read provenance data will still flag the file correctly.

For context on why that invisible layer is increasingly non-negotiable, our earlier coverage of Anthropic's C2PA watermarking for Claude outputs explains the regulatory pressure — particularly from EU AI transparency rules — pushing every major provider toward embedded provenance, visible branding or not.

How this changes a real workflow

Consider a social-media designer who batches AI image concepts in Gemini before refining them in a raster editor. Previously, every export carried the sparkle, requiring either a crop or a clone-stamp pass to clean the corner — tedious at scale. Now the toggle removes that step entirely. The same logic applies to anyone using Flow to rough out video storyboards: clean frames go straight into an edit timeline without a compositing workaround.

The music case is narrower but real. Producers using Gemini's audio generation for background tracks or sound-design sketches can now export files that don't announce their origin visually — useful when sharing rough concepts internally or embedding audio in a client presentation.

If you use Gemini for image generation, the setting is worth checking immediately: go to Gemini's settings, find Media watermark, and toggle it off. That's the entire change. The AI image generator workflow is otherwise unchanged — prompts, aspect ratios, style controls all behave identically.

The broader watermark landscape

Google's move reflects a pattern taking shape across the industry: visible branding is becoming a user preference, while invisible provenance is hardening into infrastructure. Anthropic embeds C2PA metadata in everything Claude touches — including text it only lightly edits, a scope that has caused real concern among creators. Apple is building capture-time provenance into iOS 27 camera hardware. The visible sparkle was always the soft layer; what's underneath is what regulators and platforms actually care about.

For Gemini users, the practical upshot is straightforward: cleaner outputs, same underlying accountability. Flip the toggle, keep prompting.