Short Answer
If you are usinggemini-3-pro-image, Banana Pro, or Banana 2, large differences between the generated image and the reference are usually caused by a reference-image upload format mismatch:
- These models do not support OpenAI-format reference image uploads
- References must be uploaded as base64 strings
- The issue is related to the integration method, not the model itself
Troubleshooting Steps
Walk through the following checks in order:1
Test the model in the web console
Open
imagen.apiyi.com and run the same reference through the web UI.If the web result matches the reference, the model is working correctly and the issue is in your integration. If the web result is also different, continue with the next checks.2
Check the token group
In the APIYI console, confirm your token includes the
gemini-3-pro-image or Banana-series model group.If the model is not enabled, the call may fall back to default behavior and produce unexpected results.3
Confirm the billing mode
In the token settings, check the billing mode:
- Pay-as-you-go first: uses both pay-as-you-go and subscription quota
- Pay-as-you-go only: only pay-as-you-go quota is available
4
Verify the integration method
Re-check your request against the official model documentation.For Banana 2:
docs.apiyi.com/api-capabilities/nano-banana-2-image/image-editPay attention to:- The reference field expects a base64 string, not a URL
- The MIME type must match the actual image format
- The parameter shape must match what this model supports
Common Causes
Related Questions
Nano Banana Image Failure
Common issues and fixes for Nano Banana models.
Image Async API
How to use the image async task endpoint.
How to Configure the Base URL
Connect APIYI in various client tools.
Token Model Whitelist
Configure which models a token can access.