Nano Banana 2.1 Edit API

google/nano-banana-2.1/edit
1K / 2K / 4K · 4 / 6 / 10 credits

Nano Banana 2.1 Edit uses 1–10 reference images for image blending and local edits, with consistent characters and sharp redraws. It preserves character appearance and lighting as elements change, delivering crisp results at up to 4K resolution.

Get API Key
Input
1402 / 20,000

Drag and drop files, paste from clipboard, or provide a URL. Edit requires 1–10 images. Web uploads accept JPEG, PNG, or WebP up to 30 MB each; these are uploader limits.

3/10 files added

OutputReady
6 credits / image · $0.030

Continue with

Examples

output.jpg

Use the single input illustration as the exact base. Make one precise local repair: replace only the torn crimson fabric sleeve on the fox marionette's LEFT arm, which appears on the VIEWER'S RIGHT, with neatly repaired deep teal cloth. Keep the sleeve's existing outline, length, folds, and relation to the wooden wrist. Along that sleeve's cuff add exactly three small silver crescent-moon embroidery stitches, clearly separated and readable as shapes, not letters. Preserve the other sleeve in its original red color. Preserve the fox's carved face, wooden hands and joints, all puppet strings, its cream waistcoat, the adult female puppeteer's face, gray-streaked black hair, cyan apron, her posture and hands, every major tool on the workbench, the room background, warm lamplight, shadows, camera angle, and framing. Retain the original cinematic dark-fantasy painterly style and material textures. The result should read as the same moment and same illustration with just this repaired sleeve. Do not repaint the whole puppet, alter facial identities, add objects, add text, or create advertising, brands, logos, or signatures.

output.jpg

Use image 1 as the exact base composition. Change only the time, weather, and their physically necessary lighting effects: transform this bright daytime futuristic rooftop gravity basketball court into quiet cobalt blue hour just after rain. Turn on the existing court lamps and the existing rim's restrained amber glow; do not add new light fixtures. Make the court surface damp with subtle puddles and physically believable reflections of the same player, ring, and lamps, following the existing perspective. Preserve the adult female player's exact recognizable face, short silver hair, plain coral jacket, navy trousers, copper knee brace, full shooting pose, hand position, and body proportions. Keep the single basketball at exactly the same position, the suspended hoop's size and location, all court markings, railing, skyline geometry, camera viewpoint, and framing. Keep the original cinematic retro-futurist concept-painting brushwork. Cool ambient light and small warm practical lights should alter color and shadows naturally without changing the story or moving subjects. No extra people, balls, structures, new text, advertising, brands, logos, or signatures.

Nano Banana 2.1 Edit

Nano Banana 2.1 Edit is Google's conversational image editing model for combining multiple visual references and making local changes. Built on Gemini 3.6 Flash, it accepts 1–10 reference images to extract multiple characters' appearance features and produce sharp, targeted redraws. By preserving the original composition and lighting, it provides an efficient solution for model outfit changes, scene transfers, and character design.

Why Choose Nano Banana 2.1 Edit?

  • Blend features from 1–10 reference imagesUpload up to 10 reference images at once. The model intelligently extracts and combines facial features, clothing textures, and prop materials across assets for varied edits guided by multiple images.

  • Consistent appearance across subjects and charactersTrack consistency for up to 4 characters and 10 individual objects, preserving faces and core visual features through successive edits and scene changes.

  • Sharp, precise redraws and local editsPrecisely redraw or modify selected areas, keeping instructed changes focused on the target object while preserving composition and lighting balance in unedited regions.

  • Detailed edited output at up to 4KGo beyond the usual resolution limits of editing models with native 1K, 2K, and 4K output, delivering commercial detail and texture for e-commerce visuals and advertising assets.

  • Automatic aspect ratio matching and reframingThe default auto mode follows the source image's aspect ratio. You can also choose from 10 standard ratios to rearrange the composition for different publishing channels.

Parameters

ParameterRequirementDescription
promptRequired

Non-blank text, up to 20,000 characters.

image_urlsRequired

Provide 1–10 public HTTP(S) image URLs in reference order.

sizeOptional

Output aspect ratio; auto follows the reference image.

Defaultauto1:12:33:23:44:34:55:49:1616:921:9
resolutionOptional

Output resolution, which determines the generation price.

Default1K2K4K

How to Use

  1. Prepare source and reference imagesProvide 1–10 public HTTP(S) image URLs as the base image to edit, character appearance references, or source assets for clothing and props.

  2. Write specific editing instructionsClearly describe the intended changes and the elements to preserve. A useful instruction is: "Change the specified object to the target style and keep everything else unchanged."

  3. Set the aspect ratio and output resolutionChoose auto to follow the reference image's aspect ratio or select one of 10 common ratios, then choose 1K, 2K, or 4K resolution.

  4. Submit an asynchronous editing requestCall the unified submission endpoint to start the editing task and receive a unique task_id for backend queuing and feature blending.

  5. Poll and download the finished editPoll the status endpoint until the task is ready, then retrieve the processed image for review or distribution.

Pricing

One output image per request, billed by resolution. 1 credit = $0.005.

UsageRateDetails
1K4 credits / image$0.020 / image
2K6 credits / image$0.030 / image
4K10 credits / image$0.050 / image

Best Use Cases

  • E-commerce model outfit changes and virtual try-onCombine a model's base photo with clothing detail images to dress the model in the target outfit while keeping the face, skin tone, and pose highly consistent.

  • Product background replacement and lifestyle scenesSeamlessly move products photographed against plain studio backgrounds into modern kitchens, outdoor courtyards, or cafes, with naturally integrated lighting and shadows.

  • Storyboards and new actions for multiple charactersUpload references for several existing characters and create new interactions and props while preserving each character's recognizable identity.

  • Local retouching and detail recoloringQuickly adjust details such as a sofa's color, add a specific logo, or replace accessories, avoiding complex manual selections and cutouts.

Pro Tips

  • State what to change and what to preserve: Define the edit boundaries explicitly, for example, "Replace only the umbrella in the person's hand with a folding briefcase; keep their pose, clothing, and the street background unchanged."
  • Explain the role of each reference image: When providing multiple images, identify each one's purpose in the prompt, such as "Use the person's appearance from the first image and dress them in the leather jacket from the second image," to help the multimodal encoder align features accurately.
  • Use auto for a natural composition: For local retouching or subject replacement, keeping size set to auto preserves the source image's aspect ratio and compositional space, helping avoid unexpected distortion.
  • Choose clear reference subjects: Use reference images with complete subjects, clean backgrounds, and clear lighting so the model can more accurately separate the subject from surrounding clutter.

Notes

  • image_urls is a required image array containing 1–10 publicly accessible HTTP(S) image URLs. Web uploads are limited to 30 MB per image and accept JPEG, PNG, and WebP.
  • prompt is required and must clearly describe the intended changes and preservation requirements, with a maximum of 20,000 characters.
  • size defaults to auto, following the reference image's aspect ratio. You can also explicitly select one of 10 standard ratios to reframe the image.
  • Editing tasks are billed by output resolution: 4 credits for 1K, 6 credits for 2K, and 10 credits for 4K. The number of reference images does not incur an additional charge.

Nano Banana 2.1 Edit API frequently asked questions

What is the Nano Banana 2.1 Edit API?

Nano Banana 2.1 Edit is Google's conversational model for blending features from multiple images and editing images. It combines 1–10 reference images with natural language instructions to produce high-resolution 1K, 2K, and 4K output, supporting consistency across multiple characters and sharp, targeted local redraws. Built on Gemini 3.6 Flash's native multimodal reasoning architecture, it preserves character appearance and environmental lighting when changing specific props or backgrounds, with no additional charge for more reference images. You can call it programmatically through the API or try it in the playground above.

How many reference images can Nano Banana 2.1 Edit accept per request?

The endpoint accepts 1–10 publicly accessible HTTP(S) image URLs. You can provide a base portrait, clothing detail images, and background references together. The model intelligently extracts and blends visual features across these assets.

How does Nano Banana 2.1 Edit preserve characters and composition during local edits?

The model includes subject and character consistency tracking for up to 4 characters and 10 individual objects. With prompts such as "Change the specified element and keep everything else unchanged," it preserves faces, poses, and environmental lighting in areas that are not being edited.

Does Nano Banana 2.1 Edit charge extra for multiple reference images?

No. Whether you provide 1 reference image or the maximum of 10, each generation is billed only by output resolution: 4 credits for 1K, 6 credits for 2K, and 10 credits for 4K. There is no additional reference image charge.

Which common editing tasks does Nano Banana 2.1 Edit support?

The model handles tasks such as e-commerce model outfit changes, hairstyle and clothing modifications, product background replacement with lifestyle scenes, adding or removing specific props, adjusting colors and materials, and adding new elements that match the image's perspective.

How do I configure the output aspect ratio for Nano Banana 2.1 Edit?

size defaults to auto, which follows the reference input image's aspect ratio. You can also explicitly choose one of 10 standard ratios, including 1:1, 9:16, and 16:9, to suit different publishing platforms.

How should I write effective editing prompts for Nano Banana 2.1 Edit?

Use clear, structured instructions, for example, "Replace only the person's jacket with a long, dark gray wool coat; keep their face, hairstyle, pose, and indoor background unchanged." When providing multiple references, specify the role of each image in the prompt for the best results.