Nano Banana 2.1 - Model Card

Open original

This is Google DeepMind's technical documentation for Nano Banana 2.1, an image generation and editing model based on Gemini 3.6 Flash that handles text and images with up to 1M token context. The card details its capabilities, known limitations, safety evaluations, and performance benchmarks.

  • Nano Banana 2.1 handles text-to-image generation and image editing with a 1M token input window, 4K token image output, and 64K token text output, evaluated against public benchmarks and internal regression tests.
  • The model has documented weaknesses: poor small text rendering, inconsistent character appearance between input and output, spatial confusion, and limited 3D reasoning and factuality compared to larger models.
  • Safety red teaming found similar or improved performance versus Gemini 3 Flash on child safety and content policies, with no egregious concerns; frontier safety assessment indicates no tracked or critical capability level risks.

Nano Banana 2.1 is Google's latest image generation model built on top of Gemini 3.6 Flash, designed to handle both text and image inputs up to 1 million tokens. It can output 4K resolution images and up to 64K tokens of text. The model is available through Google's API channels—AI Studio, Gemini API, and the Enterprise Agent Platform—with no special hardware requirements needed to use it. The real-world knowledge cutoff is March 2026, though some domains only have data through January 2025. The model's strengths lean toward professional image work: creating and editing images with precision, generating clear text for posters and diagrams, and handling long-context requests with localized text rendering across multiple languages. However, it has real gaps. Small text rendering often comes out blurry. Character consistency between input and output images isn't reliable. It struggles with spatial reasoning (left versus right), struggles with complex 3D reasoning, and sometimes fails to follow instructions in doodle-based editing. The model can also hallucinate like other large language models, and occasionally times out. Google's safety team conducted red teaming outside the development group and found performance on child safety met launch requirements, matching or beating Gemini 3 Flash. They assessed frontier safety risks by comparing it to Gemini 3.1 Pro and 3.7 Flash, finding no new dangerous capabilities emerged. The model didn't reach any Tracked or Critical Capability Levels, so they're confident the same holds for Nano Banana 2.1. Most technical details—architecture, training data, hardware specs—just point back to the Gemini 3.6 Flash model card rather than explaining them here.

Ask about this article