Google GeminiのAI生成コンテンツから「見えるウォーターマーク」をオフにできるようになったYou can now turn off Google Gemini’s visible watermarks

Google has begun letting users switch off the visible "sparkle" watermark that normally appears on images, videos, and music generated with its AI tools. The change, applied across the Gemini app and Google's AI video generator Flow, gives creators more control over how their AI-made outputs look and how freely they can use them, while raising familiar questions about how audiences will identify machine-generated media.

According to the update, a new "Media watermark" setting can be toggled off inside both Gemini and Flow. When it is disabled, Google removes the small sparkle mark that has been overlaid on generated content to signal its AI origin. The setting appears to be user-controlled, meaning the visible label is no longer forced onto every output by default for those who choose to turn it off. For people producing marketing assets, social clips, concept art, or music demos, the removal of a persistent on-screen badge is a practical convenience, since the mark could otherwise interfere with a clean final image or a polished video frame.

It is worth distinguishing between the visible watermark and the invisible provenance signals that Google has promoted separately. The company's SynthID technology embeds imperceptible markers directly into generated pixels, audio, or video, and it is designed to survive common edits such as cropping, compression, or color changes. Google has not indicated that turning off the visible sparkle removes these underlying signals, so SynthID-style tagging likely continues to operate even when the on-screen label is hidden. In other words, the update appears to change what viewers can see rather than eliminating the machine-readable trail that detection tools may rely on, though the practical strength of such invisible marking varies by format and by how heavily a file is later modified.

The move sits within a broader industry effort to balance creative flexibility against transparency about synthetic media. Rival systems take different approaches: OpenAI and Adobe have backed the Content Credentials framework built on the C2PA standard, which attaches metadata describing how a file was made and edited, while other platforms have experimented with visible labels, disclosure prompts, or platform-level tags applied when uploads are detected as AI-generated. Visible watermarks are the most obvious signal for ordinary viewers but also the easiest to crop out or edit away, which is part of why companies increasingly pair them with embedded or metadata-based methods that are harder to strip.

Flow, Google's AI video tool, is a notable place for this setting to surface because generated video is one of the more scrutinized categories of synthetic media, given concerns about realistic clips being mistaken for authentic footage. Extending the toggle to music generation is also significant, as audio watermarking has become a growing focus amid disputes over AI-produced songs and voice cloning. By letting creators drop the visible mark across all three formats, Google is treating the sparkle badge as an optional presentation choice rather than a mandatory disclosure, shifting more responsibility for labeling onto users and downstream platforms.

The change may prove convenient for professional and hobbyist creators alike, but it also underscores an unresolved tension. Regulators in several regions have been weighing rules that would require clear labeling of AI-generated content, and some proposals lean toward visible, human-noticeable disclosures rather than hidden metadata. If visible marks become easier to remove, platforms and policymakers may lean more heavily on invisible watermarking, detection systems, and provenance metadata to fill the gap, none of which is foolproof.

For now, the practical takeaway is straightforward. Users of Gemini and Flow who prefer output without the sparkle overlay can disable the "Media watermark" setting, and Google will strip the visible mark from the images, videos, and music they generate. Whether Google keeps embedding invisible signals in that content, and how reliably those signals hold up once files are edited or re-uploaded elsewhere, will likely shape how much this convenience affects the wider goal of keeping AI-generated media identifiable. As the tools spread, the balance between creator freedom and audience transparency is likely to remain a moving target.