Nano Banana thinking mode and thought images on Sume

Thinking mode is described on Google's page for its Gemini 3 image models. Sume's image request lists no thinking field, so you cannot set or see it; a field a model does not list returns 400 unsupported_parameter. On Sume you are billed for the completed image at the endpoint price.

Google's description is from its image generation page; Sume's billing and parameter rules are from Image generation and editing, read 2026-09-30.

What does Google say thinking mode does?

The page says the model uses a thinking process to reason through complex prompts. It generates interim thought images, visible in the backend but not charged, to refine the composition before producing the final output.

What does Sume bill?

Is there a Sume setting for more reasoning?

Not for Nano Banana. Sume's catalog advertises quality only for models that have it, and Nano Banana has none; see the quality parameter. Check supported_parameters for the model on the discovery route before sending anything extra.

What should I do for a complex prompt?

Write the composition out in the prompt, generate with n above 1 to get candidates, and pick one. For what a failed request costs, see failed image generation charges.