← Back
Replicate adds fallback model for Nano Banana Pro to handle rate limits
· featureapi · replicate.com ↗

Fallback Behavior

Replicate has introduced an automatic fallback mechanism for the Nano Banana Pro image generation model. When Google's underlying API reaches capacity and would normally return a rate limit error, the system can now attempt to generate the image using ByteDance Seedream 5.0 lite as a backup.

How to Enable

To use this feature, set allow_fallback_model to true when calling the API. The fallback is disabled by default, so existing integrations are unaffected. When triggered, the prediction logs will still show the original Nano Banana Pro request, but you can identify fallback usage by checking the resolution field in the output—it will display "fallback" instead of the actual resolution value.

Important Limitations

The fallback model has several constraints compared to Nano Banana Pro:

  • 1K resolution: Seedream 5.0 lite downscales from 2K instead of supporting native 1K
  • 4K resolution: Not supported; requests will return the original rate limit error
  • Aspect ratios: The 4:5 and 5:4 aspect ratios are unsupported; requests using these ratios will fail with the rate limit error

Pricing

When the fallback is triggered, you're charged at the fallback model's cost rate, not Nano Banana Pro's pricing. This allows you to maintain availability during peak demand at potentially different cost implications.