Regression: forced visible STAR watermark breaks Gemini Nano Banana Pro image-to-image and Flow frame-to-video workflows

Gemini API
,
1

I’ve filed a public IssueTracker bug documenting a recent regression where a forced visible STAR / sparkle watermark is injected into images generated via Gemini image models (including Nano Banana Pro).

This watermark breaks advanced creative pipelines that rely on image-to-image scene expansion and Google Flow frame-to-video interpolation. Flow interprets the watermark as a semantic object, causing it to persist, animate, or distort across frames, breaking visual continuity.

This behavior did not exist prior to the recent watermark rollout. Previously, only invisible provenance (e.g., SynthID) was present and did not interfere with workflows.

Public IssueTracker bug:

Posting here to see if other developers or creators are experiencing the same regression and to provide additional signal for the Gemini / generative media teams.

Regards,
Zarnab Khan Lashari

3

Hi @Zarnab_Lashari, welcome to the community!

To analyze the issue, could you please help us out with the exact prompt which you have given the model to generate the image?

Thank you!

5

the prompt is irrelevant. The watermark is added no matter what to every single image generated.

6

Hi @lacy @Zarnab_Lashari the watermark appears only when Nano Banana is used to generate images on AI studio. You can link a paid API key to the Nano Banana generations on AI studio to remove the visible watermark. Hope this helps!