Why Google Puts a Watermark on Gemini Images (And How to Remove It)

If you have generated an image with Gemini, you have seen it: a small four-point sparkle in the bottom-right corner of every output. It is the same logo, in the same position, at one of two sizes (48×48 or 96×96), on every Gemini, Imagen, and Nano Banana export worldwide. The badge has been there since Gemini's public launch and stayed through several model updates.

So why does Google put it there? The short answer is layered transparency. The longer answer involves a regulatory shift, an embarrassing leak from March 2025, and a quiet acknowledgment that visible badges are easier to defeat than invisible ones.

The original rationale

When Gemini first shipped image generation in 2024, the visible badge was presented as a transparency measure. The argument: people needed a quick, glanceable signal that an image was AI-generated. In an era of deepfakes, voice clones, and synthetic content flooding social networks, even a small logo in the corner was better than nothing.

That argument had real merit. Audiences who would never run a SynthID check could at least notice the badge and adjust their trust accordingly. The visible watermark was the cheapest, fastest layer of the provenance stack — not the strongest, but the most universal.

Why it never worked as intended

The visible badge was trivially defeatable from day one. Crop, blur, inpaint, or regenerate the corner — any of those neutralizes the signal in seconds. In March 2025, Google's own Gemini 2.0 Flash model was discovered to strip watermarks from copyrighted Getty Images photos with disturbing accuracy, replacing the visible mark with an "edited with AI" SynthID stamp instead. The story made headlines and embarrassed the company, which suddenly had a strong incentive to make the visible badge less central.

Beyond the security-theatre problem, the badge created a competitive disadvantage. Other major generators — OpenAI's GPT-4o, Meta's models, Midjourney on paid plans — shipped clean output by default. Professional users generating client mockups, ad creative, and presentations found that Gemini's badge read as a vendor logo on a deliverable, and they switched.

The three-layer provenance stack

The visible badge was always just one of three layers Google embeds in a Gemini image. Underneath it sit SynthID (an invisible signal distributed across every pixel) and C2PA Content Credentials (a signed metadata block). The two deeper layers are much harder to defeat, which is why Google was comfortable making the visible one optional in August 2026.

Read the full breakdown in our SynthID explainer.

What changes after August 14, 2026

Users can now turn the visible badge off inside Gemini and Google Flow via a new Media Watermark setting in their account preferences. Existing files are not retroactively modified. Free and paid plans in most countries get the toggle. India, South Korea, and Vietnam gate it to Google AI Ultra subscribers. The toggle is not available in jurisdictions that legally mandate visible AI labels.

For users with files generated before the toggle, or files exported from third-party tools that do not honor the setting, the corner badge still has to come off by hand. That is the job this site was built for.

How to remove the visible badge

Drop your Gemini image or video into the tool. Reverse alpha blending identifies the watermark at the bundled 48×48 or 96×96 size, applies the mathematical inverse of the stamp, and produces a clean copy at the original resolution. The whole thing runs in your browser in under a second per image.

Two things to keep in mind. First, the visible badge is the only layer this tool removes — SynthID and C2PA stay embedded. Second, removing the visible badge does not change your obligations under disclosure rules: if you publish AI-generated content, you still need to disclose it as AI-generated, regardless of whether the corner badge is present.