Gemini adds a toggle
to hide the visible mark
Google has added a switch that lets creators turn off the visible watermark baked into every AI-generated image, video, and song from Gemini. The invisible SynthID signal and C2PA provenance metadata stay exactly as they were — the provenance system itself hasn't changed.
Why the mark couldn't
be turned off — until now
The provenance system never changed. What was stuck was only the look of it.
On May 20, 2026, Google rolled out a provenance system for Gemini's image, video, and music generation that combines C2PA (Coalition for Content Provenance and Authenticity) content credentials with SynthID, Google DeepMind's watermarking technology. From that point on, a small translucent logo was permanently burned into the corner of every output, making it instantly obvious the content was AI-generated. Creators complained the visible mark made it awkward to post straight to social media, but Google kept it mandatory as a misuse safeguard.
On August 18, Google announced it had added a toggle to the export screen in the Gemini app and gemini.google.com that lets people turn the visible mark on or off. Turning it off doesn't touch the invisible SynthID signal embedded in the pixels or the C2PA metadata manifest that records the generating model and timestamp — both are still applied automatically. Priya Raghunathan, the Google DeepMind product lead overseeing Gemini's generative media stack, explained in a same-day technical blog post that "the visible mark and the provenance record always served different purposes. Making the former optional lets us serve creators' practical frustration and keep the provenance trail intact for misuse prevention at the same time." The change is covered in more depth by GIGAZINE's report and by TechBuzz's analysis of the impact on creator workflows.
Labeling AI output is largely driven by regulation across the industry — the EU AI Act and U.S. executive orders effectively push generators toward disclosure, which is why Gemini kept the visible mark mandatory in the first place. This change doesn't loosen that obligation; it shifts how the obligation is met, from "always visible" to "always invisible, visible by choice."
The change, in numbers
The detection-accuracy figure comes from an internal Google audit run in early August across roughly 100,000 generated samples; SynthID's detection accuracy held at 99.9% regardless of whether the visible mark was on or off. The C2PA metadata manifest is unchanged too — it still records the generating model, timestamp, and even the state of the visible-mark toggle itself, all under signature.
How the workflow changes
Generate
Create an image, video, or track as usual. The invisible SynthID watermark and C2PA metadata are still embedded automatically behind the scenes — that part hasn't changed.
Choose visibility
A new toggle on the export screen lets you switch the corner watermark on or off right there, before you save or share.
Export and share
Post straight to social or your portfolio. Because the invisible provenance data still rides along, C2PA-aware platforms and SynthID detection tools can still trace where the file came from.
Who's affected, and how
Designers, marketers, and product teams feel this differently — and should weigh it differently.
Designers
Client deliverables and portfolio pieces no longer carry a stray logo mark. Some engagements will still call for disclosure — leaving the toggle on remains an option whenever a client needs the AI origin to stay visible.
Marketers
Social ads and key visuals no longer risk showing a watermark, so more assets can go straight to submission. Creative review no longer needs a separate step to paint out the mark either.
Product teams
If you're embedding the Gemini API in your own product, Vertex AI and API support isn't due until Aug. 25, so plan around visible-mark output being the default until then.
What changes when you flip the toggle
| Visible ON (before) | Visible OFF (new) |
|---|---|
| Translucent logo in every corner | No logo shown |
| SynthID invisible watermark: applied | Applied (unchanged) |
| C2PA metadata: applied | Applied (unchanged) |
| How it looks on social | Indistinguishable from a regular post |
Video creator Jonas Meyer, a Berlin-based YouTuber with 340,000 subscribers, put it this way when TechBuzz asked him about the change.
The mark can disappear,
but where it came from doesn't. People need to keep that straight.
What to do about it
There's a real tension here. With the visible mark gone from the frame, some voices on social media are already worrying that AI-generated content will be harder for casual viewers to spot. C2PA metadata is only readable if the viewing app supports it, so an ordinary user scrolling a feed loses a way to judge authenticity by eye. SynthID detection is likewise invisible to anyone not running Google's own detection tool, so the gap between "machine-traceable" and "human-visible" arguably widens with this change, even though nothing about the underlying safeguard weakened.
For teams deciding how to use this, three concrete calls stand out. First, keep the visible mark on for client work, journalism, or anywhere disclosure matters. Second, turn it off for social-first content, but add AI-disclosure language to the post copy as a standing team policy. Third, teams integrating Gemini via the API or Vertex AI should confirm the Aug. 25 rollout has actually landed in their region before finalizing how they handle the visible mark.