From the original Nano Banana to the latest Nano Banana 2 — how Google's AI image generation has evolved.
Google's AI image generation has evolved rapidly through three distinct models in under a year. The original Nano Banana (August 2025), built on Gemini 2.5 Flash, was Google's first widely available image model - fast and affordable at 10 credits, but limited to 1K resolution with no editing or reference image support. It marked Google's entry into the AI image generation race alongside FLUX and Midjourney.
Three months later, Nano Banana Pro (November 2025) arrived as a major upgrade. Built on Gemini 3 Pro, it introduced 4K resolution, prompt-driven image editing, support for up to 4 reference images, and significantly improved text rendering across multiple languages. The trade-off was speed - Pro is slower than the original - but the quality jump was substantial, with richer detail and better prompt adherence.
Nano Banana 2 (February 2026) is the latest generation, built on Gemini 3.1 Flash. It combines the speed of the original with capabilities that surpass Pro: up to 14 reference images, multi-resolution output from 0.5K to 4K, web search context for generation, and WebP output support. It's the most capable Google image model available today.
If you've seen the name Imagen used for Google's image models - Imagen, Imagen 2, Imagen 3 - those are the same underlying technology. Imagen is the internal research name Google DeepMind uses for their image generation models, while Nano Banana is the product name used in Gemini and third-party APIs. Imagen 3 powers Nano Banana Pro, and Imagen 3.1 powers Nano Banana 2. Same models, different branding depending on whether you're reading a research paper or using the API.
Updated March 2026
Start with Nano Banana (10 credits) for the best value. When quality matters more than cost, step up to Nano Banana Pro (15 credits). All 3 models are available on Melies — try each with the same prompt and compare.
All generated with the same prompt and seed 42. Click models to toggle.
Same prompt rendered at different resolutions with Nano Banana 2.
Center crop at native resolution — notice the detail difference.
| Model | Released ↓ | Cost | Speed | Resolution | Refs | Edit |
|---|---|---|---|---|---|---|
| Feb 2026 | 8–40 | Fast | Up to 4K | 14 | ||
| Nov 2025 | 15 | Medium | Up to 4K | 4 | ||
| Aug 2025 | 10 | Fast | 1K | — |
Google's latest image model with exceptional text rendering and multi-image compositing.
Google DeepMind's premium image model with exceptional prompt adherence, rich detail, and vibrant colors.
Google DeepMind's first image generation model. Fast and affordable with solid quality for rapid iteration.
At 10 credits, Nano Banana gives you the most generations per plan. Rapid iteration, budget-friendly generation, quick concepts.
Nano Banana 2 at 15 credits delivers the highest quality. Text rendering, character consistency, multi-image compositing.
Nano Banana Pro supports prompt-driven image editing — refine and transform images without starting from scratch.
Supports up to 14 reference images to maintain visual consistency across generations.





Nano Banana, Nano Banana Pro, Nano Banana 2 and more — all in one workspace. Switch models with one click, compare results side by side. Free credits included.