Google · IMAGE

Nano Banana 2 API

Google Nano Banana 2: the current Gemini image model — strong realism, text and edits at 1K, 2K or 4K, with up to 14 reference images.

Available through Sorceress API

Nano Banana 2 model artwork

Using Nano Banana 2 in your workflow

Sorceress brings Google models into a shared API catalog. Select this model explicitly, check its supported inputs, and keep your API key on your server. Model capabilities differ; the schema and options on this page describe this entry rather than every model from the same brand.

  • Create visual concepts and iterate on art direction.
  • Build images into an asset pipeline using the supported inputs below.

Pricing and availability

9 credits at default settings. Successful work is billed in Sorceress credits. Defaults and optional inputs can affect the total.

Resolution

1K (default)9 credits

2K12 credits

4K17 credits

Inputs and supported options

modelstring · required

Model key. One of: gpt-image-2.5-flare, gpt-image-2.5-sunburst, gpt-image-2, gpt-image-1.5, nano-banana-2-lite, grok-imagine-2, grok-imagine, nano-banana-2-1, nano-banana-2, nano-banana-pro, nano-banana, seedream-5-pro, seedream-5-lite, seedream-4.5, wan-2.7-image, wan-2.7-image-pro, flux-2-pro, z-image.

promptstring · required

Text description of the image to generate.

aspectRatiostring · optional

Aspect ratio (e.g. 16:9, 1:1, 9:16). Must be supported by the chosen model.

paramsobject · optional

Model-specific options (e.g. quality, resolution, size). Affects credit cost.

refImagesarray · optional

Optional reference image URLs (for models that support image input).

API request template

Replace placeholders and choose valid values from the input contract.

POST /api/v1/tools/image_generate
Authorization: Bearer $SORCERESS_API_KEY
Content-Type: application/json

{
  "prompt": "<prompt>",
  "model": "nano-banana-2"
}
Read the endpoint reference →

Completion, results and billing

Save the job identifier returned by asynchronous requests and poll the jobs endpoint until the job succeeds or fails. Do not resubmit a generation just because it is still processing. Chat returns a completed reply and usage details. A multi-step workflow can include separately billed successful steps even if a later step fails. Keep returned asset links with your project, and consult each action for its output format.

Coding agents can discover the current catalog and input requirements from the Sorceress agent instructions. Use a Sorceress key; provider credentials stay on the server.