Introduction
Nano Banana 2 Text to image is Google's Gemini 3.1 Flash Image, released on 26 February 2026, generating images from a written prompt. Google describes it as Pro-level image quality at Flash speed. It renders at four size tiers from 0.5k to 4k in fourteen shapes up to 8:1, and it can search the web and reference pictures before drawing, starting well below Google's list price.
Choose a mode by starting material
| What you have | Mode |
|---|---|
| a written brief | Nano Banana 2 Text to image |
| an image to change | Nano Banana 2 Edit |
Key features & highlights
Economy route and Official fallback
Requests start on an economy route priced well below Google's list price. With Official fallback on, the default, a request it cannot complete runs again on the official route at the official price. Either way the model is Google's Gemini 3.1 Flash Image, with the size tiers Google publishes for Nano Banana 2.
Four tiers, fourteen shapes
From 512-pixel drafts to 4096-pixel finals, and from 1:8 towers to 8:1 banners. The shape never changes the price; only the size tier does. At 2k our 4:1 banner came back 4128 pixels wide.
Search before drawing
Web search pulls current facts such as dates, weather and landmarks; image search pulls reference pictures of real places and products. Both are optional switches on the request.
Request parameters
| Parameter | Required | Description |
|---|---|---|
prompt | Yes | Up to 5,000 characters. Describe subject, setting, light and lens in plain sentences; words in quotes are rendered as text in the image. at least 1 characters · at most 5000 characters |
aspect_ratio | No | One of fourteen shapes, including the extreme banners 4:1, 1:4, 8:1 and 1:8; 16:9 by default. The shape does not change the price. |
resolution | No | 0.5k, 1k (default), 2k or 4k. A square frame is 512, 1024, 2048 or 4096 pixels across; each tier has its own price. |
enable_web_search | No | Off by default. When on, the model looks up current facts such as weather, dates or landmarks before drawing. Default: |
enable_image_search | No | Off by default. When on, the model looks up reference pictures on the web before drawing, useful for real places and products. Default: |
official_fallback | No | On by default. Each request runs on the economy route first. Default: |
Pricing
Billed per image by size tier; the aspect ratio does not change the price. Each request starts on the economy route at the economy price. With Official fallback on, the default, a request the economy route refuses or cannot complete runs again on the official route at the official price, so the quote shows both the starting price and the most the request can cost; turn it off to pay only the economy price. Requests at 0.5k or with web search or image search run on the official route, and each search adds a charge. On the official route, a refusal that comes back as a short text reply instead of an image counts as a completed generation and is billed at the official price; other refusals are not charged.
| Specification | Price |
|---|---|
| $0.036 / image |
| $0.0384 / image (With Official fallback: up to $0.0536) |
| $0.0602 / image (With Official fallback: up to $0.0808) |
| $0.0902 / image (With Official fallback: up to $0.1208) |
Use cases
Website headers and billboards
Ultra-wide 4:1 and 8:1 banners rendered as one continuous photograph, with no stitching or stretching afterwards.
Fast fashion and lifestyle drafts
Many directions per hour at 0.5k and 1k, with the strongest re-run at 2k for delivery.
Real places and products
With image search on, scenes set at a real landmark or featuring a real product can draw on reference pictures from the web.
Tall frames for mobile
Tall 1:4 and 1:8 frames for story formats and side banners, generated as one continuous image.
Prompt guide & pro tips
Give wide frames content at both ends
For 4:1 and 8:1, describe what fills the left and right edges: a long beach, a skyline, a row of shopfronts. Otherwise the ends turn into empty sky.
Use search for facts, not style
Web search helps when the picture depends on something current or real. It does not improve style or composition, so leave it off otherwise.
Draft small, deliver large
Keep the prompt unchanged between 0.5k and 4k; the composition stays close, so choices made on the draft carry over.
Pick 2k for delivery
At 2k a square image is 2048 pixels across, which is enough for most screens and pages; keep 4k for print and heavy crops.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "google/nano-banana-2/text-to-image",
"input": {
"prompt": "A ceramic teapot on a linen cloth, soft window light, muted palette",
"resolution": "1k",
"aspect_ratio": "16:9",
"enable_web_search": false,
"enable_image_search": false
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.
