Nano Banana 2 release: Pro‑grade image generation at Flash speed

TL;DR

Google DeepMind announced Nano Banana 2, an image‑generation model that merges the advanced capabilities of Nano Banana Pro with the ultra‑low latency of Gemini Flash, making high‑quality, knowledge‑grounded images available instantly across Google products.

What is Nano Banana 2?

Nano Banana 2 (also called Gemini 3.1 Flash Image) is the latest state‑of‑the‑art image model from Google DeepMind. It delivers the world knowledge, reasoning, and visual fidelity of Nano Banana Pro while running at the Flash‑level speed of Gemini Flash. The model is designed for rapid editing, iteration, and real‑time grounding in web search results.

Core Technical Advances

Lightning‑fast inference

  • Nano Banana 2 inherits Gemini Flash’s low‑latency architecture, enabling near‑instant image generation and edit cycles.
  • The speed improvement makes Pro‑grade features accessible in interactive contexts such as the Gemini app and AI‑Mode Search.

Advanced world knowledge and grounding

  • The model taps into Gemini’s real‑time web knowledge base, pulling up‑to‑date information and reference images during generation.
  • This grounding improves the accuracy of subject rendering, infographic creation, and data visualizations.

Precision text handling

  • Nano Banana 2 can render legible, high‑resolution text within images, supporting marketing mock‑ups, greeting cards, and multilingual localization.
  • Text translation and localization are performed directly inside the image generation pipeline.

Enhanced creative control

  • Subject consistency – up to five characters and fourteen objects retain their appearance across a single workflow, facilitating storyboarding and narrative sequences.
  • Instruction following – the model adheres more strictly to complex prompts, reducing the need for post‑generation editing.
  • Production‑ready specs – supports aspect ratios from 512 px up to 4K resolution, covering vertical social posts to widescreen backdrops.
  • Visual fidelity upgrade – richer lighting, textures, and detail are maintained even at Flash speeds.

Demonstrated Capabilities

  • Infographics – generated a flat‑lay water‑cycle diagram with clean arrows and consistent styling.
  • Scientific visualizations – produced a triptych comparing cloud types in a high‑contrast comic style.
  • Artistic styles – rendered Museum Clos Lucé in bright‑colored Synthetic Cubism without textual overlays.
  • Localized signage – created a wildlife sign and automatically translated the text to Hindi while preserving visual layout.
  • Complex scenes – maintained identity of 14 characters and objects in a farm tableau, and storyboarded a six‑panel tree‑house story with consistent character design.

Product Rollout

Nano Banana 2 is being deployed today across several Google services:

  • Gemini app – replaces Nano Banana Pro in Fast, Thinking, and Pro modes; Pro/Ultra subscribers retain Pro via the image‑regeneration menu.
  • Google Search (AI Mode & Lens) – available on web and mobile in 141 new countries/territories and eight additional languages.
  • AI Studio & Gemini API – preview access through AI Studio and the Gemini API (pricing details provided on the API site).
  • Google Cloud (Vertex AI) – preview integration via the Gemini API.
  • Flow – set as the default image model for all Flow users at zero credit cost.
  • Google Ads – powers creative suggestions during campaign creation.

Provenance and Safety

  • SynthID + C2PA – Nano Banana 2 continues DeepMind’s provenance strategy by embedding SynthID watermarks and supporting C2PA Content Credentials, giving users transparent metadata about AI‑generated media.
  • Verification usage – Since November, SynthID verification has been used over 20 million times across languages, helping users detect AI‑generated images, video, and audio.

Why It Matters

Nano Banana 2 democratizes Pro‑level image generation by coupling high‑quality visual output with real‑time speed, enabling creators, marketers, and developers to iterate instantly while retaining factual accuracy and consistent subject representation. Its integration across Google’s ecosystem and robust provenance tools set a new benchmark for responsible, high‑performance generative AI.

Published by Google DeepMind on 26 February 2026

Sources