Gemini watermark toggle (Aug 2026): how to hide and what stays flagged

Google’s Gemini (via Google’s AI generation tools) added an option in August 2026 to turn off the visible watermark on AI creations, according to a report published by TechRadar contributor Axel Metz.
Worth knowing: even after you hide the visible mark, Gemini content can still be flagged as AI by backend watermarking/identification methods such as SynthID, according to the same TechRadar coverage by Metz.
That means the “visible” look is only one layer of provenance, and the moderation/labeling pipeline can still attach an AI indicator later—so what you should do changes from “remove the stamp” to “check how your export and sharing workflows treat AI labels.”

What changed in Gemini (and what didn’t)

Google’s Gemini update lets users disable the visible watermark shown on AI-generated creations, reported by TechRadar and attributed to contributor Axel Metz in its August 2026 write-up.
However, turning off what you can see doesn’t necessarily turn off what systems can detect. TechRadar’s description says Gemini creations can remain flagged as AI through backend methods like SynthID, even when the visible watermark is disabled.
In practice, that’s the key distinction: “visible watermark” is a presentation layer, while “AI flagged” is a classification layer.
Before you publish, it’s also important to confirm what generation mode you’re using (image, text styling, or editing), because different surfaces can apply different disclosure behaviors.

How to turn off the visible watermark in Gemini

Here’s the thing: the exact menu naming can differ by interface version and account rollout, but the workflow is usually consistent across Gemini’s generation tools.

1) Open Gemini and start a new generation

Start in Gemini, then create a fresh prompt and generation session (rather than editing an old result). That matters because watermark controls often attach at creation time.
If you’re using a Google product surface (web, mobile, or embedded Gemini tool), make sure you’re in the AI generation composer—not a viewing-only gallery screen.

2) Find the watermark control in the generation settings

Look for a Watermark toggle inside the generation settings panel. TechRadar’s coverage focuses on an option that specifically lets you turn off the visible watermark.
Worth noting: don’t confuse this with branding removal in templates. This toggle is meant to affect what people see on the output itself.

3) Generate and verify the output visually

After you disable the visible watermark setting, generate the asset and inspect it at full resolution. In many tools, preview thumbnails can be misleading, so open the final export/view state.
If the visible watermark still appears, check whether you’re exporting from the same generation session you toggled.

4) Assume AI-flagging may still happen after sharing

Even with the visible watermark off, TechRadar’s account says Gemini content can remain flagged via backend methods like SynthID. That means platforms or downstream systems may still label your content as AI.
So the right next step isn’t only “hide the mark,” it’s “test your distribution path”: upload to the same destination where labeling matters, then check how it’s displayed.

5) Document your settings for compliance and audit trails

If your workflow is for client deliverables, keep a simple record: the date, the tool surface, and that the visible watermark toggle was set to off.
If a publisher or platform later asks whether content was AI-generated, you’ll have a factual answer about what Gemini was configured to do visually.

Why content can stay “AI flagged” even when the watermark is hidden

Backend identification systems are designed to work even when the user removes what’s visibly shown. TechRadar’s write-up points to mechanisms such as SynthID persisting beyond the visible watermark toggle.
That’s why “visible-off” doesn’t translate to “invisible-to-detectors.” Detectors may use embedded signals or other provenance metadata that remains part of the file or generation trace.
It also means the governance question shifts: not “how do I erase the watermark,” but “how do I meet platform rules and disclosure expectations while using Gemini?”
For developers and workflow designers, this update is a reminder that provenance can be layered—UI choices don’t necessarily control classification outcomes, and moderation pipelines can be independent of what you see locally.
If you’re building or ope

What to do next (practical checklist)

First, repeat the generation on Gemini with the watermark toggle turned off, then confirm visually that the visible mark is gone in the final output state.
Second, treat AI-flagging as a separate outcome: publish to your target platform and observe how it labels or processes the file.
Finally, if your content is going to high-stakes channels (newsrooms, regulated advertising, or education), align your workflow with your organization’s disclosure policy, not just the UI watermark setting.
Forward-looking: as more tools add user controls over presentation, provenance and identification tech will likely keep evolving in parallel—so your best protection is verifying behavior end-to-end, not just toggles.

Related Articles

• Anthropic Watermarks Claude Text Output for EU Transparency Rules
• Anthropic Is Now Secretly Watermarking Everything Claude Writes
• Samsung Using Claude AI to Cut Chip Design Time by 30x

FAQs

Can I remove the visible watermark on Gemini and keep the content unflagged?

You can potentially turn off the visible watermark in Gemini, but TechRadar’s coverage says the content can still be flagged as AI through backend identification like SynthID. That means visual removal doesn’t guarantee unflagging.

Where exactly is the visible watermark toggle in Gemini?

It’s inside Gemini’s generation settings for the creation session—TechRadar’s report describes the feature update as an option to disable the visible watermark on AI-generated creations as of August 2026.

If Gemini hides the watermark, will SynthID still apply?

According to TechRadar’s write-up (by Axel Metz), backend methods such as SynthID can still be used even when the visible watermark is turned off, so AI identification may persist.

What should we test before publishing Gemini outputs?

Test in the same way you’ll distribute: generate with the watermark setting changed, export the final file, and check labeling behavior on the actual receiving platform.

Does this only affect images or also text outputs?

TechRadar’s coverage is tied to Gemini’s AI creation tools broadly, but menu availability and behavior can vary by generation mode. Confirm by testing your specific output type and export path.