Nano Banana 2 is Google's Gemini 3.1 Flash Image, designed for image creation that requires both rapid iteration and high-definition detail. It supports generating visuals from text and can also combine reference images to modify elements, styles, and colors, making it especially suitable for text-based posters, product scene changes, and visual series. Using nano-banana-2 on this platform enables both generation and editing tasks.
Clarify capacity, inputs and outputs, and invocation methods before selecting a model.
Native specifications describe model capabilities; this platform's dedicated endpoint uses explicit aspect ratio and resolution options, while reference images and other controls are available according to the supported scope of the selected endpoint.
Learn what nano-banana-2 can bring to your work.
The model excels at generating readable text together with visual compositions, making it suitable for menus, infographics, event posters, and charts. In prompts, you can separately specify titles, labels, hierarchy, and placement, allowing text to participate in the design rather than merely serving as decoration; before delivery, spelling, numbers, and line breaks should still be checked one by one.
After providing the original image, you can describe elements to add, remove, or replace, and adjust the overall style and color. When modifying product backgrounds, image atmosphere, or specific objects, it is recommended to also state what should be preserved, such as the subject's shape, camera angle, and composition, to help the model understand the editing boundaries and creative intent.
Nano Banana 2 emphasizes processing multiple reference images and maintaining consistency, allowing character, product, and scene assets to be used in the same image. Specifying the purpose of each reference image is clearer than broadly asking for them to be blended; when creating a visual series, you can reuse core reference assets and check character features and product appearance in each round.
Start with specific tasks to find where the model can be effective.
Enter a product image and a description of the target scene to generate desktop, indoor, or outdoor display images. Clearly preserve the packaging shape, colors, and key markings, then adjust the background, lighting, and color palette, making it suitable for product showcases and ad concept images; product size relationships and label text should still be reviewed before publishing.
Enter the event theme, complete copy, layout hierarchy, and canvas requirements to generate landscape promotional images or portrait social posters. You can first determine the subject and layout, then use reference images to modify the title or colors. The deliverable is a complete image, suitable for creative previews, and should not be considered a design file with independently editable text and layers.
Use the same character or product reference images and describe different backgrounds, actions, and camera shots to create a set of visually consistent assets. Specify which features must remain consistent each time, and continue using selected results as the basis for editing, making it suitable for gradually exploring character promotion, brand content, and story scenes.
Choose based on task complexity, input materials, and expected results.
Nano Banana 2 is suitable for everyday creation that simultaneously requires processing multiple reference images, text rendering, and high-definition output. Nano Banana 2 Lite focuses more on speed and scale, natively supports only 1K, and is not optimized for multiple reference inputs or continuous multi-round editing. If the task involves repeatedly changing scenes, revising text, or adjusting details, the balanced version better fits this workflow.
Nano Banana Pro corresponds to Gemini 3 Pro Image and is positioned for more complex visual tasks, brand consistency, and fine-grained creative control; Nano Banana 2 emphasizes a balance of speed and quality. The original Nano Banana corresponds to Gemini 2.5 Flash Image and is not an abbreviation for this model. Selection should be based on editing complexity and delivery requirements, rather than the name alone.
From a small-scale task to full integration.
Clarify the objective, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Example inputs and review methods to help you design your first trial.
Provide a product reference image and specify which parts need to be retained. Example prompt: “Keep the product outline, colors, and branding; generate a horizontal product hero image with a light gray background; place the product on the left, leave space for copy on the right, and use the title ‘Autumn Travel Light’; use soft natural lighting throughout.”
Compare against the reference image to check whether the subject and branding are consistent, verify the Chinese title character by character, and inspect product edges, lighting, and whitespace. In the next round, describe only the areas that need correction, avoiding simultaneous changes to the subject, background, and layout. Configure resolution and reference image input methods according to the selected API.
Understand output quality and capability boundaries before formal use.
Answers to common questions when using nano-banana-2.
No. Nano Banana 2 corresponds to Gemini 3.1 Flash Image, while the original Nano Banana corresponds to Gemini 2.5 Flash Image. When calling this model on this platform, explicitly specify model=nano-banana-2 to avoid confusing the service name with the specific model.
Use /nano-banana/images, specify model=nano-banana-2、action=generate、prompt, and set resolution=4K、aspect_ratio=9:16. The 4K here is a resolution tier and should not be understood to mean that all aspect ratios have the same width and height in pixels.
The dedicated endpoint uses action=edit, and provides the image URL and prompt editing instructions through image_urls. You can also use the image field of /openai/images/edits to submit a single image URL or an array, explicitly specify this model, and clearly distinguish what needs to be changed from what needs to be preserved.
The native model supports mixing up to 14 reference images, but this number does not guarantee the upload quantity for every calling endpoint. Actual use should follow the supported range of the endpoint, and specify the purpose of assets such as people, products, or backgrounds; do not directly treat the validation limit of a request array as the model capability limit.
The result image from the dedicated endpoint is located at data[].image_url, and includes task_id and trace_id; requests also support async and callback_url. To continue editing, use the selected result image URL as new input, describe the next modification, and progressively complete composition, text, and color adjustments.
Model information · Updated: 2026-10-01. For calling parameters and billing rules, see the API and pricing sections.
Start with clear goals and judge whether it fits your work based on real results.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs