HeadsUpAI

Google Moves Nano Banana Image Models to GA with 4K Support

Google moved its native image generation suite, Nano Banana, to General Availability (GA) via the Gemini API. The release includes Nano Banana 2 (gemini-3.1-flash-image) for high-speed tasks and Nano Banana Pro (gemini-3-pro-image) for professional assets. This follows the Nano Banana 2 launch that introduced these capabilities in preview.
Maximum resolution
4K
Reference image limit
Up to 14 images
Grounding sources
Google Web and Image Search
Model tiers
Nano Banana 2 and Nano Banana Pro
Availability
Gemini API and Google AI Studio

The shift to GA signals that these models are now stable for production-grade workflows. By integrating a Thinking process—a reasoning loop where the model explores compositions before rendering—the models excel at generating legible, stylized text. This reasoning capability allows for complex, multi-turn editing that maintains consistency across a project.

You can now generate images up to 4K resolution and use search grounding to create visuals based on real-time data. The API also supports reference blending for up to 14 images to preserve character identity. These models are available now through the Gemini API and Google AI Studio.

View the full update on ai.google.dev
Google AI Developers
Google AI Developers
@googleaidevs
X

🍌 Nano Banana Pro [gemini-3-pro-image] and Nano Banana 2 [gemini-3.1-flash-image] are now GA and ready for production via the Gemini API. Check out these great community examples to see the capabilities of both models in action 🧵↓ https://t.co/Gvv7AFwHu7

14retweets156likes
View on X

Still wondering? A few quick answers below.

The Nano Banana family includes three distinct models accessible via the Gemini API. Nano Banana 2 is the high-efficiency model optimized for speed and high-volume developer tasks. Nano Banana Pro is designed for professional asset production and complex reasoning. The standard Nano Banana model is built for low-latency tasks and high-volume efficiency.

Nano Banana image models use a reasoning process called Thinking to handle complex prompts. The model generates up to two interim thought images to test composition and logic before producing the final high-quality output. This feature is enabled by default in the API to refine the visual logic and composition of challenging generation requests.

Yes, these models support grounding with Google Search to generate images based on real-time data like weather forecasts, stock charts, or recent news events. Nano Banana 2 specifically adds the ability to use web images retrieved via Google Search as visual context, allowing the model to verify facts and generate accurate real-world imagery.

Nano Banana models generate 1K images by default but can output 2K and 4K resolutions. Nano Banana 2 also supports a smaller 512 resolution. Available aspect ratios include standard squares and wide formats like 16:9 and 21:9, with Nano Banana 2 adding specialized vertical and horizontal options such as 1:4, 4:1, 1:8, and 8:1.

Both models allow you to mix up to 14 reference images to guide the final output. Nano Banana 2 supports up to 10 high-fidelity objects and 4 characters for consistency. Nano Banana Pro is optimized for professional workflows, supporting up to 6 high-fidelity objects and 5 characters to maintain precise detail across complex multi-turn editing tasks.

Share this update