Google DeepMind has rolled out Gemini 3.1 Flash Image, branded internally as Nano Banana 2, a new image generation model that promises Flash-level speed with upgraded quality and reasoning.
The upgrade hits Gemini apps and developer tools today, replacing the prior Flash image model across Fast, Thinking, and Pro modes.
Faster image generation
Nano Banana 2 targets creators who need quick iterations without quality drops. It supports new aspect ratios (1:4 to 4:1) and resolutions from 512px squares to 4K landscapes, while improving text rendering for infographics, diagrams, and data viz.
Key upgrades include sharper details, better lighting, and stronger textures. plus Image Search Grounding, which pulls real-world context from text+image web data for more accurate outputs.
Pro subscribers can still manually select the higher-accuracy predecessor model when needed.
Complex prompt adherence has improved, though Google stayed vague on exact benchmarks like character/object consistency (no “5 characters + 14 objects” metrics confirmed).
SynthID scales to 20M+ checks
Google’s watermarking push continues: SynthID verification in Gemini has hit 20M+ uses since November, paired with C2PA Content Credentials for provenance tracking. Nano Banana 2 outputs carry these by default.
