- Updated: February 26, 2026
- 2 min read
Google unveils Nano‑Banana 2: On‑device AI with sub‑second 4K image generation
Google unveils Nano‑Banana 2: On‑device AI with sub‑second 4K image generation
Google has announced the launch of Nano‑Banana 2, the newest on‑device artificial intelligence model built for mobile devices. The 1.8 billion‑parameter model, also known as Gemini 3.1 Flash Image, leverages Dynamic Quantization‑Aware Training and Grouped‑Query Attention to deliver high‑efficiency inference, sub‑500 ms latency, and up to 30 fps at 512 px resolution. Most notably, Nano‑Banana 2 supports native 4K image synthesis, enabling developers to generate ultra‑high‑definition visuals directly on smartphones without relying on cloud resources.
The model introduces several key innovations:
- Subject consistency: Multi‑character scenes retain consistent appearances across frames, making it ideal for storytelling and animation.
- Thermal‑efficient architecture: Optimised for low power consumption, allowing prolonged usage on battery‑powered devices.
- Developer ecosystem: Integrated with Android AICore, the Banana‑SDK, and LoRA “Banana‑Peels” modules, giving developers fine‑grained control over model behaviour.
Google positions Nano‑Banana 2 as a catalyst for a new wave of AI‑enhanced mobile experiences, from real‑time photo editing to immersive AR applications. The model’s on‑device nature also addresses privacy concerns, as user data never leaves the device during processing.
Read the original story here. For more on our AI tools, explore AI Tools and visit our blog for related updates.