📰 Key Highlights

Google DeepMind has simultaneously released two new generative media models, targeting developers’ high-frequency creative workflows. The first is Nano Banana 2 Lite (model code gemini-3.1-flash-lite-image), positioned as the fastest and most cost-effective image generation model in the Nano Banana series. Benchmarks show text-to-image latency of just 4 seconds, with a cost of $0.034 per 1,000 images generated—ideal for development pipelines needing large volumes of quick drafts. Google recommends that developers still on the first-generation Nano Banana (gemini-2.5-flash-image) switch directly to this version, claiming immediate improvements in both latency and cost. While speed is the primary design goal, Nano Banana 2 Lite still maintains reliable prompt adherence, character consistency, and clear text rendering within images. The second model is Gemini Omni Flash, featuring video generation and multi-round conversational editing, now available on Google AI Studio, Gemini API, and Gemini Enterprise Agent Platform. Also launching are the Gemini app and Google Flow. The entire Nano Banana series is now divided into three tiers: Lite for ultra-low-latency high-traffic scenarios, Standard (Nano Banana 2) balancing quality and cost, and Pro optimized for complex professional use cases. Both models are now live and available for developers to try.

💬 JudyAI Lab Perspective

Google DeepMind just dropped two new models with rock-bottom costs and blazing-fast latency—directly targeting developers’ daily workflows. That’s a signal worth paying attention to for us AI builders.

The design logic behind Nano Banana 2 Lite is pretty clear: $0.034 per 1,000 images, 4-second generation. The goal isn’t to replace high-quality rendering—it’s to make the “quick draft →筛选→ refinement” loop stable enough to embed into product pipelines. Image generation is moving from “expensive one-off output” to “batch-consumable process material.”

What catches our eye is Google’s tiered approach: Lite / Standard / Pro. Developers pick the version that fits their use case instead of paying for quality they won’t use. This tiered thinking itself is a product strategy worth borrowing—it acknowledges that different workflows have different quality tolerances, rather than trying to one-size-fits-all.

If your product pipeline has image generation, it’s worth re-calculating the costs now. See if the Lite version can speed up your iteration cycle without increasing the budget.

📅 Source Info

🔗 Further Reading