Google's Nano Banana 2.1 image model arrives on Vercel's AI Gateway

Nano Banana 2.1 arrives on AI Gateway

Vercel has made Google's Nano Banana 2.1 image model available through its AI Gateway, according to an announcement on the Vercel blog. The release builds on earlier Nano Banana models with improvements to both image generation and editing, and Vercel highlights three areas as the focus: product recontextualization, mask- and ink-based editing, and factuality.

Where the model focuses

As Vercel describes it, product recontextualization positions an existing product within an entirely new scene, which is useful for generating varied product imagery from a single source photo. Mask- and ink-based editing is a more surgical capability: instead of regenerating a whole picture, the model alters only the region indicated by a mask or by ink annotations the user draws, leaving everything else untouched. Factuality refers to how accurately generated images depict real-world subjects — a long-standing weak point for image models when rendering specific people, products, or places.

On output quality, Vercel says the model produces coherent backgrounds, sharp lighting, intricate materials, and photorealistic skin tones, and does so at latency and cost comparable to a Flash-tier model. In practice, that means developers get close to flagship-level image results without paying flagship-level prices.

How to call it

The model is exposed under the name google/gemini-nano-banana-2.1 and can be reached in several ways:

• Through the AI SDK
• Through the OpenAI-compatible Chat Completions API
• Through the AI CLI

The AI CLI, Vercel notes, is an open-source terminal tool from Vercel Labs for generating text, images, video, and audio. Because it runs on AI Gateway, any Gateway model ID can be passed with the -m flag. Generated images are returned in result.files, and the model can also be tried directly in Vercel's model playground.

What the Gateway layer adds

AI Gateway is Vercel's unified API for calling models across providers. Beyond routing, it handles usage and cost tracking, retries, failover, and performance optimizations that Vercel says deliver higher uptime than the underlying providers offer on their own. The platform also includes built-in custom reporting, Zero Data Retention support, per-API-key budgets, and routing rules.

On pricing, Vercel states that AI Gateway matches what the provider charges, adds no markup, and takes no platform cut on inference — including on Bring Your Own Key (BYOK) requests, where users supply their own provider credentials. A full list of image models available through the Gateway is published alongside the announcement.

Why it matters

Image editing is becoming a standard feature in product and e-commerce workflows, and the capabilities in this release map directly onto those use cases. Recontextualization is essentially automated product photography, while mask- and ink-based editing lets users fix or restyle one part of an image without regenerating everything around it. The emphasis on factuality targets one of the bigger practical risks in this space: a generated image that looks plausible but misrepresents the actual product or subject.

For developers, the practical significance is integration cost. A model that can be swapped in via a single Gateway model ID, called through an OpenAI-compatible endpoint, and tested in a playground lowers the barrier to benchmarking it against alternatives. And because the Gateway passes provider pricing through without a markup, trying Nano Banana 2.1 does not come with a platform premium attached.