Nano Banana 2 Lite Makes Four-Second Image Generation Google's New Default
At $0.034 per 1K image, Google's speed-first model is aimed at the moments when iteration matters more than squeezing out the last degree of visual polish.
Image generation is becoming less about waiting for one dramatic reveal and more about moving quickly through ten imperfect ideas. Google's Nano Banana 2 Lite is built for that second kind of work.
Released on June 30 as `gemini-3.1-flash-lite-image`, the model produces a 1K text-to-image result in about four seconds and costs $0.034 per image, according to Google. It is available through Google AI Studio, the Gemini API and the Gemini Enterprise Agent Platform, with a broader rollout under way across Google's consumer products.
What changed
Google positions Nano Banana 2 Lite as the recommended replacement for the original Nano Banana, or `gemini-2.5-flash-image`. The company is not claiming that Lite is the most capable member of the family. It is assigning each tier a job: Lite for high-volume speed, Nano Banana 2 for a balance of quality and latency, and Nano Banana Pro for work where control and accuracy take priority.
That distinction makes the four-second figure more useful than a vague “faster” claim. It tells developers what the product is for: interactive previews, rapid concept exploration and pipelines where users may discard most of the generated images.
The distribution is unusually broad. Google says the model is coming to AI Mode in Search, the Gemini app, NotebookLM, Google Photos, Stitch, Flow and Google Ads. Outputs also use SynthID watermarking and can be checked through Google's verification surfaces.
Why it matters
The model's most important feature may be where it appears. Putting generation inside Search, Photos and NotebookLM turns it from a destination tool into an ambient capability—something users encounter while already doing another task.
For developers, the price and latency change the product math. A four-second, three-cent draft is cheap enough to make several alternatives before asking a user to commit. That can reshape interfaces around comparison and refinement instead of presenting a single synthetic answer as finished work.
What remains unclear
The performance and quality descriptions come from Google. The launch post says the model retains prompt adherence, character consistency and legible text, but it does not provide independent evidence that those qualities hold across languages, difficult compositions or repeated edits.
Cost also needs context. The quoted $0.034 is for a 1K image; total application cost depends on retry rates, moderation, storage and the surrounding model calls. Speed makes experimentation easier, but it can also make waste easier to hide.
Sources
Update note: Published and source-checked on 2026-07-24. Next checkpoint: independent quality testing and completion of the consumer rollout.
Sources
Drafted with AI assistance from verified source material and reviewed for factual accuracy, attribution, clarity, and label integrity.