Google announced it will now allow users to remove visible watermarks from AI-generated images, videos, and music. A new 'Media Watermark' toggle in Gemini and the AI video editor Flow lets users switch off the 'sparkle' icon that appears in the bottom-right corner of content produced by the Nano Banana, Omni, and Lyria models. Support for Search is coming soon.
The visible mark is going away for most users, but the underlying signals are not. According to Josh Woodward, Google's VP for Gemini, Google Labs, and AI Studio, invisible SynthID watermarks and C2PA metadata will continue to be embedded in all AI-generated content regardless of the toggle's position. Users can still verify whether a piece of content is AI-generated by asking Gemini or Search directly.
The setting will not roll out in countries where visible watermarks are legally required. That carve-out covers the EU, which recently announced new AI transparency rules that mandate visible watermarks on 'authentic-looking' deepfake content, according to Engadget. South Korea made a similar move earlier this year.
Google is also open-sourcing a new library called Credentio to let developers embed a local validation mechanism in their own apps, per TechCrunch.
The move aligns Google with the rest of the industry. OpenAI and Meta already ship AI-generated content without visible watermarks, relying instead on SynthID, C2PA, or Meta's own Content Seal standard. Anthropic this week announced it is adding invisible C2PA-based watermarks to text and files generated by Claude, partly to comply with EU regulations. Music platform Suno also announced new 'transparency tools' that label AI-generated songs when they appear on other platforms.
Woodward framed the change as a tradeoff: 'We're striking a balance here between creative control and safety,' he wrote on X.
Here's the thing. Visible watermarks were never a technical lock — they were a social nudge, and users were already editing them out. Google is simply acknowledging reality rather than pretending a sparkle icon was doing serious provenance work. The real question is whether invisible standards like SynthID and C2PA will ever reach the kind of mainstream adoption that makes them useful outside a developer portal. Right now, verifying AI content requires a deliberate extra step that most people will never take. That gap between what the metadata promises and what ordinary users actually experience is where misinformation lives — and no changelog entry closes it.
