Google is rolling out a setting that lets you turn off the visible Gemini AI watermark on images, videos and music you generate in its apps. The change was announced on 14 August 2026 and reaches users over the following few days. In India, though, the toggle comes with a condition Google mentions only in a support document: you need a paid AI Ultra subscription to see it at all.
Until now, anything generated in the Gemini app carried a small “sparkle” mark in the corner of the image, a persistent overlay on video, and a mark on the visual player generated alongside a music track. That mark is what becomes optional. Google’s own help page is explicit that the setting “only controls the visible watermark” and does not affect SynthID or Content Credentials.
What stays embedded when you turn the Gemini AI watermark off
Two invisible provenance layers remain, and you cannot switch either off:
- SynthID — Google’s imperceptible watermark, which its help page says is included in all AI-generated media created or edited with Gemini Apps.
- Content Credentials (C2PA) — industry-standard metadata describing the content’s origin, embedded alongside the file.
Google’s suggested way to check a file after the fact is to ask Gemini or Google Search whether something is AI-generated. Josh Woodward, Google’s VP for Gemini, framed the trade-off in a post on X, as reported by TechCrunch:
We’re striking a balance here between creative control and safety: while the visible watermarks are now optional, invisible SynthID watermarks and C2PA metadata are still being used for transparency. So you can still use Gemini or Search to see if an image was AI-generated.
The balance is worth reading carefully from the receiving end rather than the creator’s. The label a human being can see at a glance — on a WhatsApp forward, a screenshot, a re-uploaded reel — is the one that just became optional. The labels that survive are the ones no human can read. C2PA is metadata sitting alongside the file, so it is gone the moment someone screenshots the image or a platform re-encodes it on upload. SynthID is built into the pixels and is designed to be more durable, but it is invisible by definition — which means checking a file now requires running it through Google’s own detector rather than simply looking at it.
Why India, South Korea and Vietnam are treated differently
Woodward’s post says the toggle applies everywhere “except in countries where it’s required by law to keep them.” Google’s help page puts it differently. It states that if you are in India, South Korea or Vietnam, you will only see the setting if you have an AI Ultra subscription — otherwise visible watermarks are applied automatically to everything you create. Work and school accounts do not get the setting anywhere.
India’s regulatory context is not hard to guess at. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 were introduced by MeitY on 10 February 2026 and came into force on 20 February 2026. They created a category called synthetically generated information (SGI) covering AI-generated or AI-altered images, video and audio — notably not text. Non-prohibited SGI must be “clearly and prominently labelled”, visual SGI must carry visual labels, and, per Freshfields’ summary of the amendments, intermediaries must not allow labels or metadata to be removed or modified. An earlier October 2025 draft that would have required a watermark covering at least 10% of the surface area was dropped from the final rules.
What Google has not explained is how a subscription tier sits inside a legal carve-out. If the visible label stays in India because the law requires it, it is not obvious why paying for AI Ultra changes that; if it stays for some other reason, Google has not said what that reason is. We are not asserting that Google is breaking any Indian rule — the amendments place obligations on intermediaries, and how they apply to a generation tool’s own output settings is exactly the kind of open question lawyers have been flagging since February. But “the watermark is mandatory here, unless you upgrade” is a strange thing for a safety feature to be, and Google should say plainly which of the two explanations applies.
Where to find the setting
Once it reaches your account, on a computer go to gemini.google.com, then Settings > Media Watermark, and turn it on or off. The choice applies to all future images, videos and music tracks, and Google says it also covers Flow, its video tool. The company says the same models are covered across the board — Nano Banana for images, Omni for video and Lyria for music — with Search support coming later. Google is also open-sourcing a library called Credentio so developers can build local credential validation into their own apps.
One limit worth noting: the setting applies only to Gemini Apps on web and mobile, not to AI features elsewhere in Google’s products, and it does not retroactively change anything you have already generated.
This lands days after Anthropic began embedding invisible watermarks in Claude’s text output with no opt-out — and the two moves rhyme more than they clash. Anthropic added a machine-readable mark where there was none and gave users no switch; Google is keeping its own unremovable machine-readable layer and making only the human-readable one optional. Both companies end up with provenance that their systems can verify and ordinary readers cannot. Google’s version hands creators a real convenience while making the general public’s job of spotting AI content harder — and in India, it charges for the privilege.
Sources: Google Gemini Help, @GeminiApp on X, Josh Woodward on X, TechCrunch, Freshfields
