Google Quietly Lets Gemini Users Peel Off Visible AI Watermarks

The News

Google has begun letting people scrub the visible watermark off pictures and clips made with its Gemini tools. A fresh "Media watermark" toggle, surfaced this week, switches off the small "sparkle" badge that until now sat in the bottom-right corner of anything produced by the company's Nano Banana image model and its Omni system. The control also reaches Flow, Google's dedicated video generator.

Switching the badge off does not make the output untraceable. Google confirmed that two invisible markers stay baked in: SynthID, its own imperceptible signal, and C2PA metadata, the cross-industry content-credentials standard. Either can be read back through Gemini or a Search query to confirm a file was machine-made. Josh Woodward, vice-president of Google Labs, Gemini and AI Studio, said the hidden layers persist even once the visible one is gone.

The change was reported on 14 August 2026. Google intends to extend the toggle to Search, while keeping the visible mark compulsory in jurisdictions whose laws require it.

Why It Matters

The visible watermark was always the compromise that let generative-image firms argue they were being responsible. Making it optional shifts the industry's honesty burden onto signals that ordinary users cannot see. Google is not alone: OpenAI, Meta and Anthropic have all leaned on invisible watermarking rather than a badge on the pixels, so this move formalises a direction the sector had already chosen.

The last time provenance felt this contested was the 2024 election cycle, when fabricated political imagery pushed platforms to promise clearer labelling. Two years on the trend has reversed: labels are hidden by preference rather than shown. That suits creators who disliked the sparkle, but it hands trust to detection tools most people never open and not every platform bothers to read.

Indian Angle

For India the timing is awkward. In late 2025 the Ministry of Electronics and Information Technology (MeitY) floated draft amendments to the IT Rules that would push platforms to label "synthetically generated information" with a prominent, durable marker, with one proposal pegging the label at no less than a tenth of a still image's surface or the opening tenth of an audio clip. A consumer-facing switch that strips exactly that kind of visible badge runs against the grain of where Indian rule-making is heading.

The gap bites hardest in fintech, where deepfake fraud is already a live problem. Both the RBI and SEBI have warned over the past year about fabricated videos of officials and bogus investment tips spreading on WhatsApp and YouTube, and SEBI has cautioned investors about doctored clips of its own leadership. Invisible SynthID and C2PA tags help forensic teams, but they do nothing for an ordinary saver scrolling a feed, who can no longer rely on a visible mark to separate a genuine notice from a synthetic one.

There is a home-grown dimension too. Indian model builders such as Sarvam and Ola-backed Krutrim ship their own generative systems, and they must soon decide whether to follow Google in making visible marks optional or hold a stricter line to stay ahead of MeitY. Whichever way the regulator lands, Indian platforms distributing Gemini output may find themselves re-imposing a badge Google has just taught users to remove.

FAQ

What exactly can users now switch off?

Only the visible "sparkle" watermark in the bottom-right corner of images and videos made with Gemini, Nano Banana, Omni and Flow. Invisible identifiers, namely Google's SynthID signal and C2PA content-credential metadata, remain embedded and cannot be toggled away by the user.

Does removing the badge make content undetectable?

No. The file still carries hidden SynthID and C2PA data. Anyone can, in principle, check origin by uploading it to Gemini or running a Search query, though this depends on the checking platform actually reading those signals, which many social feeds do not do automatically.

How does India's approach differ?

MeitY's draft IT Rules amendments lean towards mandatory, prominent visible labels on synthetic media, sized as a share of the image or audio. That is close to the opposite of an optional, user-removable badge, so Google may have to keep the mark switched on for Indian users if the rules are finalised.

Where can I read the original announcement?

The change was first reported by The Verge on 14 August 2026, drawing on Google's updated Gemini settings and comments from Google Labs vice-president Josh Woodward.

This story was reported by The Verge. Read the full original coverage at The Verge.