Google Gemini adds toggle to remove visible AI watermarks

Google announced that Gemini’s new “Media watermark” toggle lets users hide the sparkle‑style watermark on AI‑generated images, videos and music, a move that shifts the visible signal to an optional setting while preserving invisible SynthID and C2PA tags for verification [1].

Toggle mechanics and user control

The toggle appears in Settings → Media Watermark and defaults to “On,” meaning the visible watermark is shown unless the user disables it. When off, the bottom‑right sparkle disappears from content created with the Nano Banana, Omni and Lyria models. Google stresses that the invisible SynthID watermark and C2PA metadata remain embedded, allowing anyone to query Gemini or Search to confirm whether a piece of media was AI‑generated [1][2].

Industry context and transparency implications

Google’s decision mirrors practices at OpenAI and Meta, which already rely on invisible watermarks without mandatory visible tags. Anthropic has similarly added invisible marks to its text and image outputs. By making the visible label optional, Google aims to balance creator flexibility with ongoing transparency obligations under the EU AI Act, which mandates machine‑readable provenance for synthetic media. Critics note that the visible watermark was the most immediate cue for casual users; its removal could make AI‑generated content harder to spot without dedicated verification tools [3][4].

Competitive landscape

Competitors continue to experiment with watermark strategies. OpenAI’s DALL‑E and Meta’s image generators embed only invisible SynthID/C2PA signals, while Anthropic’s recent watermark rollout targets compliance with EU regulations. Google’s optional watermark distinguishes its approach: it retains a user‑controlled visual cue for those who want it, while ensuring a persistent, non‑visual provenance layer across all outputs.

What to watch

• Search integration – Google plans to extend the toggle to Search, but rollout will exclude countries where visible watermarks are legally required.
• Regulatory pressure – Ongoing EU AI Act enforcement could prompt Google to adjust the default setting or enhance verification tools.
• Competitor responses – Watch for announcements from OpenAI, Meta or Anthropic on new watermarking features that could influence market expectations for AI‑generated media labeling.

The shift to an optional visible watermark underscores a broader industry move toward invisible provenance as the primary means of tracking AI‑generated content, raising questions about how effectively users and platforms will adopt the underlying verification tools.