Google shipped Nano Banana Pro to general availability in June 2026 and nobody made a big deal of it. The I/O keynote spotlight went to Gemini Omni and Managed Agents. But for anyone building an app that generates or edits images, the model formerly known as Gemini 3 Pro Image is now the most capable reasoning-driven image model with a public API — at $0.134 per 1K or 2K image, $0.24 for 4K.
The name is a Google internal codename that leaked and stuck. Nano Banana 2 (Gemini 3.1 Flash Image) is the cheaper, faster sibling. Nano Banana Pro is the high-quality lane. Both are now generally available in the Gemini API.
What Nano Banana Pro Actually Does
Most image generation models work the same way: you send a text prompt, they return pixels. Nano Banana Pro adds a layer that matters if you build anything beyond basic generation: native image editing through a joint reasoning-generation process. You don't patch pixels externally. You send the original image plus an instruction in natural language, and the model applies changes while preserving everything you didn't ask it to touch.
That sounds incremental. The specific thing it does better than the alternatives is text rendering. Accurate text inside generated images — product labels, UI mockups, infographic callouts, signage — has been an industry failure mode since the original Stable Diffusion era. Nano Banana Pro is the first model where "add the text 'Sale' in bold white on the product" reliably produces readable text rather than decorative gibberish.
