What happened
TechCrunch reported on August 14, 2026 that Google is rolling out a new setting allowing users to remove the visible watermark it normally places on AI-generated images, video, and songs. Per TechCrunch, Josh Woodward, Google's VP for Gemini, said the change applies to content generated with Nano Banana, Omni, and Lyria, and that Google is "striking a balance here between creative control and safety." The outlet reports the toggle is available now in Gemini and in Flow, Google's video editor, with support in Search described as coming soon.
According to TechCrunch, this only affects the small visible mark that appears on the file. The invisible SynthID watermark, which Google embeds directly into the pixel or audio data of a generation, and C2PA provenance metadata, an industry standard for recording a file's origin, both remain in place regardless of whether the visible watermark is on or off. LumoMate has not independently verified details beyond what TechCrunch reports; this briefing attributes claims to TechCrunch and Woodward rather than presenting them as separately confirmed.
Why it matters
The distinction between a visible watermark and invisible provenance data is easy to miss but matters for anyone trying to judge whether an image, video, or song was AI-made. A visible watermark is cosmetic: it is a label stamped on top of the content that a viewer can see, and Google's new toggle simply lets creators turn that label off. SynthID and C2PA metadata work differently. SynthID embeds a signal directly in the underlying data of the file, and C2PA metadata is a standardized record attached to the file describing its origin and edit history. Removing the visible watermark does not touch either of these.
That said, invisible provenance systems are not a guarantee. Detection tools like SynthID can be affected by heavy editing, format conversion, screenshotting, or re-uploading to platforms that strip metadata, and C2PA metadata can be lost when a file passes through software that does not preserve it. Neither system offers airtight proof that a given file is or is not AI-generated in every case, so treat both the presence and absence of a visible watermark as one signal among several, not a final verdict.
What to do next
- If you are trying to judge whether content might be AI-generated, do not rely only on the presence or absence of a visible watermark, since Google's new setting lets creators remove that label without affecting invisible provenance data.
- If your workflow depends on checking provenance, look into tools built around SynthID or C2PA metadata rather than a visual watermark, and keep in mind that neither is guaranteed to survive heavy editing or re-uploading.
- Watch for the Search rollout TechCrunch describes as coming soon, and for whether other AI platforms introduce similar visible-watermark controls.
- Read the original TechCrunch report for the full context, including Woodward's framing of the change as a balance between creative control and safety.
This briefing is based on TechCrunch's August 14, 2026 report. Claims about the toggle, Nano Banana, Omni, Lyria, SynthID, and C2PA are attributed to TechCrunch and the Google executive it quotes; LumoMate has not independently verified them beyond that reporting.