Nano-Banana-2 Image Editing Tool
Gemini 3.1 Flash Image from Google: generation and editing with up to 14 references, output from 512px to 4K, and aspect ratios as extreme as 8:1 and 1:8.
Overview
Google released Gemini 3.1 Flash Image, nicknamed Nano Banana 2, in February 2026, and its model card says it is based on Gemini 3 Flash. Google moved several Nano Banana Pro features into it: readable in-image text, translation of the text inside an image, and world knowledge for infographics and diagrams, with output up to 4K. Google also says it follows complex, multi-part prompts more strictly. It is the only Nano Banana page on Wiro with a 512px tier, and its ratios run from 1:8 to 8:1.
Choosing a variant: go with Nano Banana 2 for catalog and ad work with many object references, localized variants or extreme formats. Nano Banana Pro, built on Gemini 3 Pro, is the model Google suggests for the most complex visual tasks, and it keeps 5 people consistent against 4 here. Nano Banana is the 2025 Gemini 2.5 Flash Image release; it takes 3 inputs, renders near one megapixel and is the only sibling that exposes temperature.
What you can build
- Ad localization flows that translate the copy inside a finished ad into other languages.
- Magazine covers, labels and greeting cards with legible titles, like the "FUTURE WEAR" cover among the example prompts on this page.
- Catalog and lookbook images that keep up to 10 products faithful to their reference shots.
- Storyboards where up to 4 characters keep their look from panel to panel.
- Web banners and skyscrapers at 4:1, 8:1, 1:4 or 1:8.
Inputs
The prompt field is mandatory. Up to 14 reference images can go in as uploads or URLs (JPG, PNG, GIF, WebP or HEIC); of those, up to 10 can be objects reproduced with high fidelity and up to 4 can be characters kept consistent.
Resolution has four tiers: 512, 1K (the default), 2K and 4K. Aspect ratio can follow the input image or use one of 14 ratios: 1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9 and 21:9. A square runs from 512x512 to 4096x4096, and 8:1 at 4K produces a 12288x1536 strip.
Safety Setting offers five values, BLOCK_LOW_AND_ABOVE through OFF, and starts at OFF.
Recommended settings
- Try a prompt at 512 or 1K first, then run it at 2K or 4K. There is no seed input, so the 2K or 4K run is a fresh image and will differ from the draft.
- When text matters, quote the exact words and name the target language for translations.
- For 1:8 and 8:1, describe how content runs along the long edge, for example a row of products from left to right.
Limitations
The Gemini 3.1 Flash Image model card names these weak spots:
- Tiny lettering tends to blur, above all at 1K, and long passages or full pages of text often come out wrong.
- Characters resemble their references closely, though not always exactly.
- Instructions given through masks or doodles are only partly followed.
- Placement sometimes confuses left and right.
- The model's grasp of world facts and 3D space has limits, and its knowledge stops at January 2025.
Safety & compliance
Gemini 3.1 Flash Image marks its images with Google's SynthID watermark, hidden from view but detectable with SynthID verification. The model card refers to Google's Generative AI Prohibited Use Policy. When a reference image shows a real person, make sure that person has agreed to its use.
Example prompts
Great starting points for nano-banana-2.
API quick start
Run nano-banana-2 with a single API call.
{
"prompt": "A top-down, realistic street photography …",
"inputImage": "https://your-cdn.com/input.png",
"aspectRatio": "1:1",
"resolution": "1K"
}curl -X POST "https://api.wiro.ai/v1/Run/google/nano-banana-2" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "A top-down, realistic street photography …",
"inputImage": "https://your-cdn.com/input.png",
"aspectRatio": "1:1",
"resolution": "1K"
}
JSONOther models you might like
Explore similar image generation models.
