Google Will Let You Turn Off Watermarks on Its AI-Made Images, Videos, and Songs

Google will soon let people switch off the visible watermark on images, videos, and songs created by its AI tools. The option is rolling out in the coming days across Gemini and Google's Flow video editor, with support for Google Search following soon after. TechCrunch

Josh Woodward, Google's VP for Gemini, said the toggle will cover the Nano Banana, Omni, and Lyria models. Users can turn the visible watermark on or off by going to Settings > Media Watermark.

The important part: turning off the visible watermark does not remove the invisible watermark Google also embeds. Google uses a system called SynthID, which adds a hidden digital mark to AI-generated images, videos, and audio without changing how they look or sound. A tool called SynthID Detector can check whether an image was created or edited using Google AI, and the Gemini app uses SynthID to verify an image's origin too.

Think of it like a visible price sticker on a product, and also a barcode printed underneath. You can peel off the sticker, but the barcode is still there if you know how to scan it.

Google is also releasing a free, open-source tool called Credentio. It lets other developers add a way to check whether digital media is authentic — essentially verifying where a piece of content came from — directly inside their own apps, without needing Google's products.

The broader context here involves several threads. At Google I/O 2026, Google announced it was expanding SynthID, and OpenAI said it would adopt the same watermarking standard. Both companies, along with others, originally pledged to watermark AI content back in July 2023 at a White House announcement. Google is also keeping the visible watermark — a small sparkle icon — on images made by free-tier and Google AI Pro-tier users of Nano Banana Pro, so the new toggle does not apply to everyone equally.

More recently, Anthropic announced it would add a watermark to text and files generated by Claude to comply with EU regulations. Google's decision to make its visible watermark optional goes in the opposite direction, relying more on the invisible SynthID layer as the main way to track where content came from.

The tension is between giving users control and making sure people can tell when something is AI-generated. A visible watermark is simple: anyone scrolling through a social feed can see at a glance that an image was made by AI. SynthID, being invisible, requires special tools to detect. Ars Technica reported in late July that simply cropping a SynthID-watermarked image can often remove the hidden mark. If users turn off the visible watermark and the invisible one can be removed by something as basic as cropping, the system has a real gap.

In my view, the bet Google is making is clear. By open-sourcing Credentio and getting other companies like OpenAI to use SynthID, Google is pushing for invisible, behind-the-scenes tracking as the industry standard instead of visible labels. The question is whether invisible watermarks can hold up in the real world, where images get cropped, re-saved, and passed through platforms that routinely strip out hidden data.

Google's announcement also comes alongside a busy product cycle. The company announced Gemini 3.7 Flash, which it calls its most capable model yet for coding and AI agents, and published its Made by Google 2026 Pixel device announcements. Together, the watermark toggle, Credentio, and new models suggest Google is pulling its AI tools into a more complete platform.

The practical takeaway is that Google is acknowledging something that has been true since visible watermarks were introduced: they get in the way for creators who want clean, polished content, and they are easy to remove from the image anyway. Relying more on SynthID and C2PA metadata while letting users control the visible mark makes sense, as long as the invisible layer actually survives everyday actions like cropping and re-saving. That is a big "as long as," and future testing of how well SynthID holds up will tell us whether this approach works or is just more convenient.