I’ve been experimenting with the image model a lot of folks in the Google/Gemini ecosystem casually call “Nano Banana 2” (aka Nano Banana Pro), and built a small web playground around it:

https://www.nanobananapro.site

Nano Banana Pro is a next‑generation image model focused on higher fidelity and better reasoning. Compared to the earlier “Nano Banana” image API, it aims to improve in a few specific areas:

- Native 2K output with clean 4K upscaling - Sharper micro‑details and more realistic materials - Much more stable text rendering (labels, UI, posters, etc.) - Intent‑driven composition for complex prompts - Flexible aspect ratios (1:1, 4:5, 16:9, 21:9, 9:16, …) - Better character identity and style consistency across generations - Stronger inpainting/outpainting with scene‑aware edits

The site is meant as a simple playground for:

- Testing prompt engineering for complex scenes (multiple characters, actions, and constraints in one prompt) - Trying typography and layout (banners, UI mockups, posters with multi‑line text) - Exploring editing workflows like masking, extending a scene, or refining parts of an image - Comparing how it handles physics/spatial logic vs. other image models you’ve used

I’m particularly interested in feedback from:

- People building production‑grade creative tools or UGC pipelines (avatars, covers, marketing visuals) - Folks who care about text rendering and layout quality - Anyone who’s been hitting limits with current image models on aspect ratios, consistency, or editing

Questions I’d love feedback on:

- What’s missing for this to be useful in a real workflow? - Which knobs (guidance, composition controls, aspect‑ratio presets, editing tools) would you want exposed? - What would make it easier to plug something like this into your own product or pipeline?

If you try it and manage to break it (or find cases where it fails badly), I’d really appreciate examples and thoughts.