Google just handed creators a long-requested option. Users can now switch off the visible sparkle logo that has marked images, videos and audio from its Gemini models for years. The change arrives amid a flood of third-party tools that already erase those marks with a few clicks. Yet the company insists its deeper protections remain untouched.
The announcement landed Friday. Josh Woodward, VP for Gemini, Labs and AI Studio, spelled it out on X. “We’re striking a balance here between creative control and safety,” he wrote. A new setting in Gemini and the Flow video editor lets users toggle visible watermarks on or off. Support for Search comes soon. The feature targets Nano Banana, Omni and Lyria generations. It rolls out in the coming days.
From Annoyance to Optional Feature
That sparkle in the bottom-right corner grated on professionals. Marketing teams, designers and social media managers complained it branded their output too obviously as machine-made. Client presentations looked tainted. Stock imagery uploads triggered automatic flags. So Google listened.
But don’t mistake this for surrender on provenance. Invisible SynthID signals and C2PA metadata stay embedded no matter what. Ask Gemini or Search “Is this made with AI?” and the system still detects the origin. Woodward emphasized the point. “That means you can still verify if an image has one of these markers by asking Gemini or Search if it’s AI-generated.”
The Verge first detailed the policy shift today. Its report notes the toggle won’t apply in countries that mandate visible marks under local rules. OpenAI, Meta and Anthropic have leaned harder on invisible methods or content credentials from the start. Google now aligns closer to that approach while giving users choice.
Yet the timing feels reactive. For months online tools have promised one-click removal of Gemini’s visible badge. Sites like geminiwatermark.io process files locally in the browser. No upload. Original quality preserved. GitHub repositories host open-source scripts using reverse alpha blending, a mathematical trick that subtracts the semi-transparent overlay without degrading the pixels beneath. Chrome extensions do the same inside the browser.
These solutions exploded in popularity after Google introduced the watermark. Early versions targeted the simple logo. Later ones handled video frames across entire clips from Veo or Omni. Developers updated detection logic within days whenever Google tweaked logo placement or opacity. The arms race moved fast.
And Google itself once showcased similar capabilities. Back in March 2025, users discovered Gemini 2.0 Flash excelled at erasing watermarks from stock photos. TechCrunch reported how the model didn’t just delete Getty Images stamps. It filled the gaps with convincing background detail. One X user declared, “New skill unlocked: Gemini 2 Flash model is really awesome at removing watermarks in images!” Another noted the model replaced the old mark with its own subtle indicator.
Google responded then with a warning. A spokesperson said using its tools for copyright infringement violates terms of service. The company would monitor closely. That episode highlighted a deeper tension. Visible marks deter casual misuse but prove trivial to strip. Invisible ones offer better long-term defense yet require active verification.
Industry watchers see this week’s move as pragmatic evolution. Creators want clean assets for commercial work. Agencies hesitate to deliver files sporting an obvious AI tag. At the same time, regulators and platforms push for reliable signals against deepfakes and misinformation. SynthID, which Google has deployed across billions of pieces of content, addresses that need. Recent I/O announcements expanded its verification inside the Gemini app, Search and Chrome.
Critics still question effectiveness. Last April a developer claimed to reverse-engineer SynthID using only 200 generated black and white images plus signal processing. The Verge covered those assertions. Google pushed back, arguing the method didn’t truly defeat the system. Detectors still worked in most tests. The episode showed how determined actors can probe these defenses.
Similar stories play out across the sector. Samsung once added watermarks to AI-edited Galaxy photos only to see users erase them with the phone’s own object remover tool. OpenAI warned that C2PA metadata often gets stripped by social platforms or screenshots. The signals break easily in the wild.
So where does Google’s decision leave content authenticity? Better than before, some argue. The visible toggle gives legitimate users breathing room without fully abandoning detection. Professionals can generate clean files for legitimate purposes while SynthID lingers for platforms and researchers. Verification tools in Search and the Gemini app make checking easy.
But the proliferation of removal tools suggests visible marks never offered strong protection anyway. Anyone motivated enough could scrub them. The real safeguard lies in those hidden markers and the ecosystem built around them. Google open-sourced its Credentio library to help developers adopt similar standards.
Legal risks persist. Removing watermarks from copyrighted stock images without permission still counts as infringement under U.S. law. Getty and others have pursued cases before. Terms of service violations can lead to account bans. Google says it continues monitoring.
For now the policy reflects a maturing market. AI generation has moved from novelty to daily production tool. Watermarks that once signaled experimentation now feel like an outdated stamp on professional output. Giving users the off switch acknowledges that shift. Keeping SynthID intact shows the company hasn’t given up on provenance.
Expect the conversation to continue. As models improve and adoption grows, pressure will mount for standards that balance creator freedom with public trust. Google’s latest step offers one version of that balance. Others may follow. The sparkle logo, long a fixture in generated images, may soon become a choice rather than a requirement. The invisible signals, however, look set to endure.