Google will peel back a new era of AI images with Nano Banana 2 (www.techradar.com)

Google has quietly previewed Nano Banana 2 (internally GEMPIX 2), a follow‑up image generator that’s appearing in the Gemini app and other Google experiments. Leaked previews and code references show the model delivers cleaner lines, sharper angles, better color precision and improved on‑image text editing. Google is also surfacing a higher‑end “Nano Banana Pro” in commits, suggesting a premium, high‑resolution tier is in the works. The big technical shift is behavioral: Nano Banana 2 adopts a multi‑step, self‑correcting workflow—planning an image, rendering a draft, analyzing for mistakes, fixing them, and iterating until the output meets its own criteria. That internal critique loop mirrors human design workflows and represents a move from reactive single‑pass generation to proactive refinement. For the AI/ML community this matters because it demonstrates a scalable approach to reducing common generative flaws (composition, perspective, text legibility) without manual post‑editing, and it signals broader adoption of iterative architectures and internal evaluation modules in image models. Wider rollout across Google surfaces also means these advances could quickly propagate into consumer and pro creative tools, raising expectations for image fidelity and controllability in generative systems.