Google launches Nano Banana 2, bringing pro-level image generation to its Flash model

Google DeepMind today released Nano Banana 2, its latest image generation and editing model.^[1] Officially designated Gemini 3.1 Flash Image, the model is designed to deliver the quality and reasoning capabilities of Google's higher-end Nano Banana Pro at the faster speeds associated with its Flash tier.^[1] It is rolling out immediately as the default image model across the Gemini app, Google Search, Flow, and Google Ads.^[1]

Background

The original Nano Banana — technically Gemini 2.5 Flash Image — launched in August 2025 and became a viral phenomenon, particularly in India, where it drove millions of image generations through the Gemini app.^[1] Google followed it in November with Nano Banana Pro (Gemini 3 Pro Image), which offered studio-quality output and more advanced creative controls but ran on the slower, more expensive Pro model.^[6]

Nano Banana 2 attempts to collapse that trade-off. The model brings several capabilities that were previously exclusive to the Pro tier while running at Flash speeds, effectively replacing Nano Banana Pro as the default across most Google surfaces.^[1]

What's New

The model introduces what Google calls "advanced world knowledge." It draws on Gemini's real-world knowledge base and pulls real-time information and images from web search to render subjects more accurately.^[1] Google demonstrated this with a developer demo called "Window Seat," which generates photorealistic views from any window location in the world using live local weather data at up to 4K resolution.^[2]

Other notable capabilities include subject consistency across up to five characters and 14 objects in a single workflow, which enables storyboarding and narrative sequences without characters drifting in appearance between frames.^[1] The model also offers improved precision in text rendering and in-image translation, allowing users to generate legible text for marketing materials or greeting cards and then localize that text into other languages directly within the image.^[1]

Resolution support ranges from 512px to 4K across a variety of aspect ratios.^[3] Google says the model delivers sharper details, richer textures, and more vibrant lighting compared to the original Nano Banana.^[1]

For developers, the model introduces configurable thinking levels — minimal by default, with a high/dynamic option that lets the model reason through complex prompts before rendering, improving output quality for sophisticated requests.^[2]

Where It's Available

Nano Banana 2 is rolling out today across a broad set of Google products.^[1] In the Gemini app, it replaces Nano Banana Pro as the default across Fast, Thinking, and Pro modes.^[1] Google AI Pro and Ultra subscribers retain access to Nano Banana Pro for specialized tasks through the three-dot regeneration menu.^[1]

In Google Search, Nano Banana 2 becomes the default for AI Mode and Lens across 141 countries, available through the Google app and on mobile and desktop browsers.^[1] It is also the new default image generation model in Flow, Google's video editing tool, where it is available at zero credits.^[1] The model powers image suggestions in Google Ads as well.^[1]

For developers, it is available in preview through the Gemini API, Gemini CLI, AI Studio, Vertex AI, and Google's Antigravity development tool.^[3] A new templates feature in the Gemini app lets users pick from preset artistic styles — including monochrome, gothic clay, steampunk, and others — before generating.^[1]

Benchmark Performance

Independent evaluation appears to support Google's claims. Artificial Analysis ranked Nano Banana 2 first in text-to-image generation and third in image editing at roughly half the price of its closest competitors, according to reporting from OfficeChai.^[5] If accurate, that positioning represents a significant value proposition for developers weighing quality against cost at scale.

Early hands-on testing by The New Stack found the model noticeably faster than its predecessor, though it still occasionally produced errors in logo rendering and text accuracy in certain prompts.^[8]

Safety and Provenance

All images generated by Nano Banana 2 carry SynthID watermarks, Google's technology for identifying AI-generated content.^[3] The company is also coupling SynthID with C2PA Content Credentials, the cross-industry standard for content authenticity metadata developed by a coalition that includes Adobe, Microsoft, Google, OpenAI, and Meta.^[3]

Google says its SynthID verification feature in the Gemini app has been used over 20 million times since launching in November, helping users identify AI-generated images, video, and audio.^[1] C2PA verification is coming to the Gemini app soon, which will provide users with richer context about not just whether AI was used to create content, but how.^[1]

Competitive Context

The launch comes at a moment when the AI image generation market is shifting from novelty toward production deployment. Enterprise customers are increasingly looking for models that balance quality, speed, cost, and compliance tooling — rather than simply the model that produces the most visually impressive output.^[7]

On one flank, open-weight alternatives like Alibaba's Qwen-Image-2.0 are pushing toward comparable quality without per-image API charges, though they lack the built-in provenance infrastructure that regulated industries increasingly require.^[7] On the other, Google's own Nano Banana Pro remains available for use cases demanding maximum fidelity.^[1]

Nano Banana 2 is positioned squarely in the middle — fast enough for high-volume production workflows, capable enough to handle complex creative tasks, and priced aggressively enough to undercut slower Pro-tier alternatives from competitors.^[5]

Why This Matters

Google has effectively collapsed two product tiers into one. Features that previously required the more expensive, slower Pro model — world knowledge, subject consistency, advanced text rendering — are now available at Flash speeds across Google's consumer and developer platforms.^[1][4]

For individual users, the practical effect is better image generation in the tools they already use, at no additional cost. For developers and enterprises, the calculus is more consequential: the speed-versus-quality trade-off that has defined AI image generation workflows is narrowing,^[7] and the embedded provenance tooling addresses a compliance requirement that self-hosted alternatives cannot easily match.^[7]

Whether Nano Banana 2 delivers on these promises at production scale remains to be seen. Early third-party benchmarks are encouraging,^[5] but real-world performance under diverse workloads is a different test. Independent evaluations over the coming weeks will determine whether Google has genuinely closed the gap or simply repackaged the marketing.