Google Lets Gemini Users Hide Visible Watermarks While Keeping Invisible AI Provenance Markers

Google is giving users a way to remove visible watermarks from media generated through Gemini and Flow. The setting applies to AI-created images, videos, and music, according to announcements and reporting published on August 14 and 15.

The change separates the branding people see from the provenance signals embedded in generated files. Coverage from TechCrunch and The Verge identifies the feature as a new “Media watermark” setting, while the headlines from Thurrott and Engadget describe the same shift as a way to hide Gemini marks on generated media.

Turning off the visible label won't remove Google's invisible identification systems. SynthID remains embedded in supported AI-generated content, and C2PA-related metadata also stays attached, TechCrunch reported. The Verge likewise said Google will continue placing invisible SynthID and C2PA markers in images, videos, and music.

Those systems serve a different purpose from a visible logo. A visible watermark signals an item's origin to viewers, while invisible markers help platforms and other tools inspect provenance later. Removing the first therefore doesn't eliminate Google's underlying content-identification measures.

Visible watermarks can interfere with media intended for presentations, creative drafts, social posts, or other polished layouts. Google’s setting gives Gemini and Flow users more control over how generated work appears when it leaves the app.

The change also reflects a practical tension in generative media: people want clear disclosure, but they often dislike branding that dominates an image or video. Android Authority’s headline points to a related question around Gemini’s on-device Nano experience, although its available excerpt provides no details confirming the same rollout. Watermark-removal websites also appear in search results, but their listings provide no verified information about Google’s feature.

Google’s approach keeps machine-readable provenance in place while making the disclosure less visible to ordinary viewers. That design gives creators a cleaner output, but it also makes visual inspection less immediate for audiences who encounter a file outside Gemini or Flow.

SynthID and C2PA markers work only when receiving services, applications, or investigators know how to detect and preserve them. Files can also pass through editing and publishing workflows that affect metadata. The reported policy therefore preserves technical signals without guaranteeing that every viewer will know a piece of media was generated by AI.

The “Media watermark” setting is coming to Gemini and Flow, with the rollout described as a Google policy and product change rather than a third-party removal technique. The available reporting does not establish a single global launch date or confirm identical controls across every account, device, and media type.

Users should distinguish Google’s built-in option from independent watermark-removal services. The latter are promoted by several low-reliability pages, but the available listings contain no substantive reporting about Google’s implementation. The key change is Google’s own control: visible marks can be hidden while SynthID and C2PA-related provenance data remain part of the generated file.