I’m experiencing inconsistent performance with gemini-3-pro-image-preview (Nano Banana Pro) when using multiple reference images.
Issue:
Until late-February, gemini-3-pro-image-preview was reliably generating high-quality outputs from multiple reference images. However, since the launch of Gemini 3.1 Flash Image (Nano Banana 2), the Pro model has become highly inconsistent:
∙ When it works: Outputs are excellent—accurate detail reproduction, good aesthetic quality
∙ When it fails: The model appears to downsample reference images so aggressively that it can’t see or replicate fine details, resulting in generic or inaccurate outputs
The inconsistency makes it impossible to rely on for production use. I’ve tested gemini-3.1-flash-image-preview as a backup, but while it’s more consistent, the outputs have an overly glossy, plasticky aesthetic that doesn’t meet our quality standards.
Questions:
- Is there a known infrastructure or resource allocation issue with gemini-3-pro-image-preview following the 3.1 Flash launch?
- Are there updated best practices for maintaining consistent quality when using multiple reference images with the Pro model?
- Is there a timeline for stabilization of the Pro model’s performance?
Any guidance, workarounds, or timeline information would be extremely helpful.