Nano Banana 2 supports a set of predefined aspect ratios rather than arbitrary pixel dimensions. The supported ratios include 1:1 (square), 4:3, 3:4, 16:9, and 9:16, which cover the most common use cases for web, mobile, and video content. You specify the desired aspect ratio in the generation request using a string parameter, and the model generates an image at its native resolution for that ratio. For 16:9, the native output is 1344×768 pixels; for 9:16, it is 768×1344 pixels. Requesting an unsupported ratio will return a validation error with a message listing the accepted values.
The choice of aspect ratio affects more than just the dimensions of the output image. The model's composition behavior adapts to the canvas shape, so a 16:9 request for a landscape scene will tend to distribute elements horizontally in a way that takes advantage of the wide format, while the same prompt at 9:16 will reframe the scene vertically. This means you may need to adjust your prompt phrasing for different ratios if you want a specific compositional outcome—for example, explicitly stating that the subject should occupy the center third of the frame for portrait orientation, rather than assuming the model will infer an appropriate framing from context.
If your application requires an aspect ratio not in the supported set—for example, a 21:9 ultrawide format for a desktop banner or a custom square crop with a non-standard pixel count—the practical approach is to generate at the nearest supported ratio and then crop or letterbox the output using an image processing library. Post-generation cropping is lossless from a quality perspective and gives you precise control over the final dimensions without requiring a custom model configuration.