Google launches Nano Banana Pro image model built on Gemini 3 with flawless text rendering
Nov 20, 2025
Key Points
- Google launches Nano Banana Pro, a Gemini 3-based image model that solves the text rendering problem plaguing prior generative models by eliminating character errors and doubled letters.
- The model generates complex infographics and interactive UI elements with precision control, positioning it for enterprise and document-heavy use cases rather than simple image generation.
- Nano Banana Pro still struggles with semantic reasoning, failing to hide objects in camouflage tasks and instead placing them prominently in image centers.
Summary
Google launched Nano Banana Pro, a new image generation and editing model built on Gemini 3. The model renders text without the typos and character errors that plagued earlier versions, generates complex infographics with precision control, and handles advanced world knowledge tasks. Sundar Pichai announced the model offers state-of-the-art performance for both generation and editing with emphasis on text precision.
The model generates interactive UI elements such as scrollable visualizations around research reports, positioning it for complex document-heavy use cases rather than simple image generation. One limitation: it struggles with "Where's Waldo" puzzles, placing Waldo prominently in the center rather than hiding it in a crowd, suggesting it has not internalized the semantic logic of camouflage.
The competitive significance lies in the text rendering breakthrough. Prior image models would occasionally fail with doubled letters or character errors even when text was visually convincing. Eliminating that class of error lowers friction for enterprise and professional use cases where text accuracy in generated images matters.