Nano Banana Pro API on Sume: 4K tiers, aspect ratios, and edits

To call the Nano Banana Pro API on Sume, send POST /v1/images with model: "google/nano-banana-pro" (or "google/nano-banana-2"), a prompt, and optional resolution, aspect_ratio, and input_references fields. Both models generate from text or edit from up to 10 reference images, render at 512, 1K, 2K, or 4K, and take no quality parameter.

The details come from the Image API docs, the Nano Banana note on the Image 1.0 page, and the catalog at GET /v1/images/models, read 2026-09-26.

Which Nano Banana model ids can I send?

Send google/nano-banana-pro (catalog name Nano Banana Pro) or google/nano-banana-2 (Nano Banana 2); the bare nano-banana-pro and nano-banana-2 work as aliases.

The two share one descriptor set with a single exception: Nano Banana 2 adds 4:1, 1:4, 8:1, and 1:8 to the ratio list. Both are edit-capable; image generation API models compares them with the rest of the catalog.

Which resolutions can I pick, and what is the default?

resolution takes 512, 1K, 2K, and 4K on both models, and 1K applies when you omit it. The older spelling 0.5K still works for 512. size is shorthand for the same tiers, so size: "4K" works too.

These two are the only catalog models with a 4K tier. A 4K render is also more likely to outlast the 30-second wait, so handle a 202 job envelope as well as a 200; 4K, quality, and image count has the cross-model view.

How do ratios and pixel sizes work on Nano Banana?

auto lets the provider pick the ratio. Nano Banana takes no custom pixels: a WIDTHxHEIGHT or { width, height } value in image_size maps to a native ratio, so 1080×1350 becomes 4:5. Nano Banana Pro renders aspect_ratio: "4:5" at about 928×1152 at 1K.

Here is how the two models differ:

How do I edit an image with Nano Banana Pro?

Put up to 10 public HTTPS image URLs in input_references and send aspect_ratio: "auto", which both models list, so the output follows the reference. The shared reference rules are in image generation with reference images.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-pro",
    "prompt": "place this product on a marble counter in morning light",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/product.png" } }
    ],
    "aspect_ratio": "auto",
    "resolution": "2K"
  }'

What does Nano Banana reject?

A parameter a model does not list is rejected with 400 unsupported_parameter rather than silently dropped. On Nano Banana, these requests fail:

• Any quality value, because neither model advertises the field.
• mask_url or background, which only the ChatGPT Image 2.5 ids list.
• Pixels on size. Put them in image_size, where they map to a ratio.
• An n above 4. Both models cap n at 4, and a larger value is refused with a 400 error.

How is Nano Banana billed?

Per image, at a rate that depends on the tier. The catalog pricing line shows the 1K rate. Nano Banana 2's estimate changes for 512, 2K, and 4K; Nano Banana Pro's changes only at 4K, which costs more than its default. Catalog lines include Sume's margin, plus a 5.5% agent fee by default; see API pricing.