TechCrunch: Google removes visible watermarks from AI-generated images and video

On August 14, 2026, TechCrunch reported that Google is giving users the option to remove visible watermarks from content generated by its AI tools. The change applies to Gemini, the Flow video editor, and a set of image and audio models including Nano Banana, Omni, and Lyria. Invisible watermarks through the SynthID system remain in place.

The shift is framed around professional use. Josh Woodward, VP of Google Labs, acknowledged that visible watermarks “can often make content less useful for professional and creative work.” The decision recognises that designers and creative professionals who use AI generation as part of a production workflow — producing visual concepts, editing assets, generating reference imagery — have been constrained by marks that are appropriate for public-facing content but disruptive at intermediate stages.

SynthID, the underlying detection system, operates differently from visible marks. It embeds identification data within the file itself through subtle, imperceptible changes to the generated output. The watermark is detectable by Google’s own tools and will be accessible through Gemini and Search, but does not affect how the image looks or how it can be used. C2PA metadata documenting AI origin also remains embedded.

For designers, the practical outcome is cleaner outputs when AI generation is one step in a larger workflow. A rough concept generated in Gemini, passed through illustration tools, or combined with original photography no longer carries a visible mark that signals its origin at every stage. The invisible tracking remains, which means provenance can still be verified for accountability purposes — but visible disclosure becomes opt-in rather than a default.

The change reflects a broader tension in how AI content labelling is being implemented across platforms. The EU AI Act’s Article 50 requirements (also taking effect this month) push toward more mandatory disclosure, while Google’s move gives professionals more control over how AI origins appear in their outputs. The two positions are not necessarily in conflict: SynthID satisfies machine-readable provenance requirements while the visual mark is left to the creator’s discretion.