Nano Banana 2 Lite
Google released Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image), described as the fastest and cheapest Gemini image model for high-volume use. Simon Willison tested it with a "Where's Waldo"-style prompt and found the results better than previous Nano Banana models, though the model made spelling errors in the generated image.
Background
Gemini 3.1 Flash Lite Image is Google DeepMind's new budget-tier image generation model, branded "Nano Banana 2 Lite" in AI enthusiast shorthand. It is positioned as the fastest and cheapest option in Google's Gemini image model lineup, optimized for high-volume, low-cost use cases.
- Simon Willison is a well-known UK-based developer and AI blogger who regularly tests new models and shares his findings.
- The "Where's Waldo" (also known as "Where's Wally") test is a popular informal benchmark for image generators: it checks whether a model can produce a complex, cluttered scene with many characters and reliably place a specific object (here, a raccoon with a ham radio) in it, testing compositional reasoning and prompt fidelity.
- Willison previously tested this same prompt on other "Nano Banana" models (OpenAI's GPT Image 2 variants) in April; he finds Google's new model produces a more satisfying result, though it still makes basic spelling errors in the image.
- The "Nano Banana" nickname for Google's image models originated from an earlier Google model (Gemini 2.0 Flash Image) and has stuck as community jargon.