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 images from text and can also modify elements, styles, and colors using reference images. It is 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.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
Input parameters and result formats vary by service. Use the public API for this model and follow its guide for generation, task retrieval and editing operations.
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.
The model excels at generating readable text together with visual composition, making it suitable for menus, infographics, event posters, and charts. In prompts, specify titles, labels, hierarchy, and placement separately, so text participates in the design rather than serving only as decoration; before delivery, spelling, numbers, and line breaks should still be checked individually.
After providing the original image, you can describe elements to add, remove, or replace, as well as adjust the overall style and color. When modifying product backgrounds, image atmosphere, or specific objects, it is recommended to also specify what should be retained, such as the subject shape, camera angle, and composition, helping the model understand the boundaries of the modification and the creative intent.
Nano Banana 2 emphasizes multi-reference image processing and consistency, allowing people, products, and scene assets to be used in the same image. Assigning a purpose to each reference image is clearer than broadly asking for them to be blended; when creating visual series, you can reuse core reference assets and check character features and product appearance in each round.
Provide a product image and a description of the target scene to generate desktop, indoor, or outdoor display images. Explicitly preserve the packaging shape, colors, and key identifiers, then adjust the background, lighting, and color palette. Suitable for creating product displays and advertising candidate images; product size relationships and label text should still be reviewed before publication.
Provide 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 titles or colors. The deliverable is a complete image, suitable for creative previews, and should not be regarded as a design file with independently editable text and layers.
Use the same character or product reference image, and describe different backgrounds, actions, and 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. Suitable for the gradual exploration of character promotion, brand content, and story scenes.
Nano Banana 2 is suitable for everyday creation that simultaneously requires multiple reference image processing, text rendering, and high-resolution output. Nano Banana 2 Lite is more focused on speed and scale, natively supports only 1K, and is not optimized for multiple reference inputs or multi-round continuous editing. If the task involves repeatedly changing scenes, editing 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.
Choose generate for text-based creation; choose edit to modify existing assets, provide reference images with image_urls, and separately describe what to preserve and what to change.
Specify model=nano-banana-2, action, and prompt to /nano-banana/images; start with aspect_ratio=1:1, resolution=2K, and count=1, setting the canvas format and resolution separately.
Get images from data[].image_url; for asynchronous requests, query or receive callbacks with task_id. When continuing edits, pass in the selected image again and narrow the scope of changes in each round.
Keep the backpack in the reference image in the same shape, color, and branding, place it on the left side of a light gray tabletop, with the title “Travel Light” on the right. Make the title clear, with overall soft natural lighting.
Check the Chinese text character by character, as well as product edges and branding; continue with the selected image using edit, changing only one design requirement per round.
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, and prompt, then set resolution=4K and aspect_ratio=9:16. Here, 4K 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 between content to modify and content to preserve.
The native model supports mixing up to 14 reference images, but this number does not guarantee the upload quantity for every endpoint. In practice, follow the supported range of the endpoint and describe whether the assets are for people, products, backgrounds, or other purposes; do not treat the validation limit of a request array as the model capability limit.
Result images from the dedicated endpoint are located in data[].image_url and include task_id and trace_id; requests also support async and callback_url. To continue editing, use the image URL of the selected result as new input, describe the next modification, and progressively complete composition, text, and color adjustments.
Choose a development approach or application suited to this model and follow its setup guide.
Python SDK Integration Guide
SDK installation, authentication and examplesTypeScript SDK Integration Guide
SDK installation, authentication and examples
