Four seconds and under four cents per thousand images generated. On Tuesday, Google opened its day with the launch of Nano Banana 2 Lite, the newest member of its family of AI image generators. A calling card that speaks the language of numbers: speed and cost, for developers who need to produce visual content at scale.

The technical identity card

The model promises generation in about 4 seconds, with aggressive pricing below $0.04 per thousand requests. This is not an on-device or downloadable version: Nano Banana 2 Lite is available as a cloud API, embedded in the ecosystem of Google’s managed services. A choice that simplifies integration but raises the usual questions about where data resides and who has control over the infrastructure.

When pricing challenges on-premise logic

For those building image generation pipelines, the economic calculus is often the deciding factor. Such a low cost can shake even the most on-premise-oriented assessments: if an image costs less than a cent, the TCO of an on-premise GPU cluster – including purchase, energy consumption, and maintenance – becomes hard to justify for low or intermittent volumes. But the reasoning isn’t straightforward. Constant workloads and the need for spending predictability, along with latency and customization requirements, can still make an internal infrastructure preferable, perhaps built on open models like Stable Diffusion.

The sovereignty and control game

Image generation often touches sensitive data: from medical material to product concepts yet to be patented. In regulated industries, data flowing through a public API represents a compliance risk that no low price can zero out. Nano Banana 2 Lite, however cheap, does not offer by default data residency guarantees or isolation comparable to an on-premise deployment. This is where the reflection shifts from cost to control: those choosing the self-hosted route pay a higher hardware price, but maintain full sovereignty over information and the processing chain.

A market in ferment, a few signals

Google’s launch lands in a landscape where visual generation is becoming a commodity, accessible at costs that a year ago seemed unrealistic. For IT decision-makers, this means constantly recalibrating selection parameters. It’s no longer just about deciding between on-premise and cloud, but doing so knowing that the economic gap is narrowing. AI-RADAR will continue to track these models and offer analytical tools to navigate the options, because generation speed matters as much as clarity in architectural choice.