📰 Key Takeaways

Google announced on Friday that users will soon be able to remove the visible watermark from AI-generated content, including images, videos, and songs, though this won’t affect the hidden SynthID watermark or the metadata tied to the C2PA standard. Josh Woodward, VP of Gemini, posted on X that the toggle will apply to three models — Nano Banana, Omni, and Lyria. The option to turn off the visible watermark will roll out first on Gemini and Google’s video editing tool Flow, with support on Search coming a bit later. This shift reflects how Google is rethinking the way it labels AI-generated content: visible watermarks flag where content came from, but they also often make that content less usable for professional or creative work. So Google is trying to strike a balance between “creative freedom” and “safety and transparency” — the visible watermark becomes optional, but the hidden SynthID watermark and C2PA metadata stick around. Users can still check whether an image is AI-generated through Gemini or Search tools. The feature will roll out gradually over the coming days, and once it’s live, users will be able to toggle the visible watermark on or off through the Media Watermark option in settings. Google also open-sourced a new library called Credentio at the same time, letting developers build local verification into their own apps. This policy shift comes right after Anthropic sparked a lot of discussion by adding watermarks to text and files generated by Claude in response to EU regulations.

💬 JudyAI Lab Take

Google announced on Friday that users can now remove the visible watermark from AI-generated images, videos, and songs, while the hidden SynthID watermark and C2PA metadata stay in place. This is worth paying attention to if you’re building with AI, because it shows how platforms are calibrating the line between “content provenance” and “creative freedom.”

The change applies to three models — Nano Banana, Omni, and Lyria — rolling out first on Gemini and Flow, with Search following later. Google’s logic here: a visible watermark flags where content came from, but it also makes that content less usable for professional or creative purposes. So they’re handing the “show it or not” decision back to the user, while keeping the “can we verify it” capability baked into the system (SynthID + C2PA). This dual-track design — optional visible labeling paired with mandatory hidden verification — reflects an industry-wide shift from blanket mandatory watermarking toward layered governance. It’s a nice contrast to the discussion Anthropic stirred up recently when it added watermarks to Claude’s output in response to EU regulations. Google also open-sourced its Credentio library alongside this, letting developers build their own local verification — another way of pushing responsibility out to the ecosystem.

If you’re building a product involving generated content, worth asking yourself: is your content provenance mechanism relying on user goodwill, or is it enforced by the system?

📅 Source Info

🔗 Further Reading