Google DeepMind Introduces Vision Banana: An Instruction-Tuned Image Generator That Beats SAM 3 on Segmentation and Depth Anything V3 on Metric Depth Estimation

Google DeepMind introduced Vision Banana, a lightly instruction-tuned version of Nano Banana Pro that converts vision tasks into RGB image outputs.

The model reports state-of-the-art or near-state-of-the-art results on segmentation, depth, and surface normal benchmarks without training on those datasets.

Its performance suggests one generative model may handle both image generation and core perception tasks, reducing the need for separate specialist systems.