News

Nano Banana Pro Lands in ComfyUI: Studio 4K and the Partner-Node Era

8 September 2026 · RenderBob team

Google DeepMind's Nano Banana Pro (Gemini 3 Pro Image) is now in ComfyUI. It targets production-ready visuals: native 4K, clean text, character consistency, and a partner node on Google's infrastructure.

Google DeepMind's Nano Banana Pro (Gemini 3 Pro Image) is now available in ComfyUI, and it is a marker of where production image generation is heading. It targets production-ready visuals: native 4K generation with studio controls for colour grading, angle and focus; clean, legible text rendering with detection and translation across ten languages; blending of up to fourteen reference images; strong character and concept consistency; and world knowledge grounded in Search for accurate real-world detail. Its faster sibling, Nano Banana 2 (a Gemini 3.x flash image model), targets high-volume use.

For motion and design studios, the capabilities matter. Clean text and character consistency are exactly the things open image models have struggled with, and they are exactly what client work demands. How you run it matters more. Nano Banana Pro comes to ComfyUI as a partner node: the node calls the model over an API, billed in Comfy credits, with the option to run it through Comfy Cloud with no local setup at all. There is no checkpoint to download and no GPU requirement for that node. The compute happens on Google's infrastructure.

The partner-node era changes the shape of a workflow

Your graph is no longer necessarily models on my GPU. It is increasingly a hybrid: some nodes run open weights on local hardware, and some nodes are thin clients calling a closed model in someone else's cloud. A single workflow might sample video locally, then hand a frame to Nano Banana Pro for a 4K text-accurate treatment, then composite the result back: three very different execution locations in one graph.

Access is instant. Governance is not

You get instant access to frontier models with zero provisioning, no VRAM ceiling, no model management. The trade-offs are just as real, and studios should name them. Cost moves from capex to metered per-call spend that needs a ceiling. Your client's material now leaves your environment on those API calls, which is a governance and NDA question, not just a technical one. Reproducibility gets harder when a node's behaviour lives on a vendor's servers and can change under you. Per-model concurrency limits and pricing become a scheduling concern of their own.

Partner nodes are genuinely useful, and Nano Banana Pro is a real capability jump. Treat the hybrid graph deliberately. A production pipeline needs to know which nodes run local and which call out, what leaves the building on each API node, what each call costs, and how to keep a workflow reproducible when part of it is a remote service. Frontier models arriving as nodes raise capability. Governing where your data goes and what your renders cost is the work that turns capability into something a studio can safely sell.