Nano Banana image generation through the API

Update, October 2026. Image generation through Zukko is temporarily unavailable. For the current model list, see the Models page↗ or GET https://api.zukko.pro/v1/models. Closest replacement: none yet — only text models are currently available in Zukko.

Images live on a separate endpoint — text models go through /v1/chat/completions or /v1/messages, images through /v1/images/generations. Billing differs too: per finished image rather than per token.

The minimal request

curl https://api.zukko.pro/v1/images/generations \
  -H "Authorization: Bearer fetch_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-2-text-to-image",
    "prompt": "Microservice architecture diagram, flat graphics, light background"
  }'

The endpoint follows the OpenAI format, so a client library that already speaks image generation connects by changing the base URL.

Three modes, three models

The mode comes from the model you pick rather than a parameter — worth knowing, because it is easy to get wrong.

The images field takes an array of URLs or data URLs. For editing that is the source picture; for reference mode, a sample of style or composition.

What an image costs

The price is fixed per image; prompt length does not affect it.

The developer variants cost half. The lite line is cheaper still and noticeably faster; on simple graphics the difference in output is small, on complex scenes it shows.

The practical consequence: iterate on prompts with the lite model and do the final pass on the full one. Ten tuning iterations on lite cost 28 cents against 80.

What not to expect

Image models handle text inside a picture poorly: captions come out with errors, and that is a property of the model class rather than this particular one. If your layout needs readable text, it is safer to leave space for it and place the caption as a separate layer.

Reproducing several references at once is also hard. Reference mode carries overall style well and specific details less so.

Connecting

The key and address are the same as for text models: https://api.zukko.pro/v1 for OpenAI-compatible clients, a fetch_... key in the Authorization: Bearer header. The list of available models comes from /v1/models, while prices and limits come from /v1/models/info (or the Models page in the dashboard).