Google lets users strip the visible watermark from AI content

Google is giving users a switch to turn off the little "sparkle" icon that has marked its AI-generated images, videos, and music since Gemini's early days. Josh Woodward, who leads Google Labs, Gemini, and AI Studio, confirmed the change on August 14, 2026, framing it as a matter of user choice rather than a retreat from transparency.

The toggle lives under Settings > Media Watermark and applies across Gemini and Flow, Google's AI video generator, covering output from the Nano Banana and Omni models. Google says the rollout begins within days of the announcement.

Turning off the visible mark does not remove Google's underlying provenance system. SynthID and C2PA metadata, both invisible to the naked eye, stay embedded in every generated file regardless of the toggle's position. That distinction is doing a lot of work here: Google is separating cosmetic labeling from the forensic layer that lets platforms and researchers later verify AI origin.

The company plans to extend the same toggle to Search, though it will hold the setting back in countries where regulations mandate visible AI labels. That carve-out suggests Google is treating the visible watermark as a compliance feature in some markets and a user preference everywhere else.

Catching up with rivals

Google's sparkle watermark had made its outputs easy to spot at a glance, something OpenAI's and Meta's image tools never imposed by default. Those companies rely on invisible standards instead: SynthID and C2PA in some cases, and Meta's own Content Seal elsewhere. By making the visible mark optional, Google's defaults move closer to that norm.

Anthropic is moving in a related direction, applying invisible watermarks to both AI-generated text and images, a signal that provenance tracking is consolidating around metadata rather than on-image branding across the industry.

Google is also open-sourcing a library called Credentio, letting developers build local validation checks directly into their own apps rather than depending solely on Google's infrastructure to confirm whether content was AI-made.

The company is betting that invisible, machine-readable provenance can do the trust work that a visible logo used to signal at a glance.

Whether that bet holds depends on adoption. Invisible watermarks are only useful if enough tools and platforms actually check for them, and removing the one signal ordinary users could see without any tooling at all raises the bar for casual verification, even as it brings Google's defaults in line with the rest of the industry.