nano-banana-2

To generate images with reliable in-image typography, use subject-first instruction structures and exact text quoting in your natural-language prompts. This approach ensures more predictable text rendering and framing within the generated output.

You can control aspect ratio and resolution by configuring JSON input fields such as aspect_ratio and resolution tiers in the API request. This allows you to generate rapid draft-style images or final outputs tailored to specific dimensions.

Web grounding works for text-to-image generation by enabling an optional web search parameter. This allows the model to reference current events or real entities, providing real-world context and freshness in the generated scene.

The best way to create rapid batch variants is to specify the num_images parameter in your JSON input request. This allows you to quickly iterate and produce multiple variations of an image from a single natural-language prompt.

You set generation limits and safety tolerance by configuring the limit_generations and safety_tolerance JSON input fields. These parameters help control the volume of outputs and filter content according to your specific safety requirements.