Google has released Nano Banana 2, an updated image-generation model in the Gemini family that brings measurable gains in speed and image fidelity.

Model improvements

According to Google, the model shows improved understanding of user prompts and produces more accurate text overlays embedded within generated images.

It also delivers more consistent multi-character scenes and object relationships, reducing common compositional errors that previously affected complex outputs.

• Better prompt comprehension and clearer generation of textual elements inside images.
• Stable multi-character scenes, with improved spatial arrangement and interaction between objects and people.
• Reference selection via search to better match source imagery and stylistic cues in outputs.
• Support for up to 4K resolution outputs for higher-fidelity images.

Availability and testing

The model will be integrated into Gemini, AI Studio, Vertex AI, and Flow, making it available across Google's developer and consumer interfaces.

Google indicates that users will be able to test the model for free through supported platforms, without specifying rollout timelines or regional limits.

Nano Banana 2 replaces the prior base image generator within Gemini's toolset and represents an incremental update to Google's image generation capabilities.