Nano Banana

Google’s Gemini image-gen with native edit-mode — image-in + prompt → image-out. Flat $0.046/image. Sync API.

Overview

nano-banana is Google’s gemini-2.5-flash-image — a Gemini-family image generator with one killer feature most others lack: native edit-mode. Pass a reference image alongside the prompt and the model edits it in context (inpainting, style transfer, object swap) rather than starting from scratch.

The preview successor nano-banana-3-flash ships the newer Gemini 3.1 image stack on the same edit-mode pattern. See nano-banana-3-flash.

Specs

Field	Value
Model ID	nano-banana
Creator	Google
Backend	Vertex AI (gemini-2.5-flash-image, us-central1)
Best for	In-context image editing, style transfer, low-cost iteration
Edit mode	Yes — pass image_url to enable
Aspect ratios	Suggested in prompt (model interprets — not contract-enforced today)
Pricing mode	Flat per image (1K tier default)
Default latency	~5–10s
Output	Blob-hosted URL (Vercel CDN)

Pricing

Variant	Provider cost	Kyma list
nano-banana	$0.034	$0.046

Flat rate. Vertex’s publisher endpoint accepts an imageConfig.imageSize field but ignores it today (probed 2026-05-16) — every call returns ~1K output. When Vertex honors size control we’ll switch to per-quality tiering (low/medium/high).

Live source: GET https://kymaapi.com/v1/pricing.

Compared to other image models on Kyma

Strength	nano-banana	nano-banana-3-flash	imagen-4	flux-kontext-pro
Edit mode (image-in)	★★★★★	★★★★★	—	★★★★★
Speed	★★★★★	★★★★★	★★★★	★★★★
Cost	★★★★★	★★★★★	★★★★	★★★★
Bleeding-edge quality	★★★★	★★★★★ (preview)	★★★★	★★★★
Style transfer	★★★★	★★★★★	—	★★★★

Use this when

• You need to edit an existing image (inpaint a background, swap an object, change style).
• You’re iterating on a reference image and want each variant to be cheap.
• You’re prototyping conversational image workflows (the model accepts multi-turn editing).

Pick something else when

• You need text-from-scratch without a reference image → any other image model is fine.
• You need legible text in image → gpt-image-2.
• You need multi-reference blending (10 sources) → flux-2-pro.
• You need vector / SVG output → recraft-v4-vector.

Example — pure text-to-image

curl -X POST https://kymaapi.com/v1/images/generations \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana",
    "prompt": "A tiny yellow star on a black background",
    "n": 1
  }'

Example — edit-mode

curl -X POST https://kymaapi.com/v1/images/generations \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana",
    "prompt": "Add a small black cat in the foreground, keeping the rest of the scene unchanged",
    "image_url": "https://example.com/scene.jpg",
    "n": 1
  }'

Async endpoint — poll GET /v1/jobs/{id} until succeeded.