Since around June 18, 2026, image generation/editing with the preview image models returns heavily blurred, low-resolution output — same prompt and settings that worked fine the day before.
-
Affected: gemini-3-pro-image-preview, gemini-3.1-flash-image-preview
-
Not affected: gemini-2.5-flash-image works correctly; OpenAI works fine
-
Reproduces in Google AI Studio directly (not an SDK/client issue)
-
Possibly related: generationConfig.imageConfig.imageSize (1K/2K/4K) appears to be ignored — output comes back at ~1K regardless of the requested size (see js-genai #1461, litellm #18656)
Is this a known regression, and is a fix in progress? Thanks
Fyi here’s what it produces
I have similar issues. High-resolution (2K & 4K) image generation returns washed-out / semi-transparent / “ghosted” subjects with multi-reference requests on gemini-3.1-flash-image and gemini-3-pro-image — native 1K is correct.
IM Having the exact same issue trough AI STUDIO API.
similar issue here , been trying to find workarounds all day
Exacly the same problem here through a Google AI API key: references sent are ok, but render Nano Banana 2 & Nano Banana Pro are totaly blurry like painted.
Same issue, web chat wasted 30mins with Google tech support who has no idea.
Thanks for the report. This has been fixed as of Friday Jun 19th.
