google / nano-banana-pro

This is the one people mean when they say "Nano Banana" in 2026: the flagship, Gemini 3 Pro Image - 4K native output, reference-image support, and the "thinking mode" that reasons about your prompt before drawing. CivitaiImageGenGoogleNanoBananaPro wraps it for ComfyUI as a hosted node, and it's the closest thing the Civitai pack has to a flagship image node: higher ceiling, higher Buzz cost, fewer free passes.

What it is. One of the ~160 generated nodes in the official civitai/civitai-comfy-nodes pack (engine google, model nano-banana-pro). Text in, IMAGE out, zero local compute. The whole appeal of the Pro tier, per Google's own framing and the community's read of it: 4K native generation and "thinking mode" - the model spends a beat reasoning about your prompt before it generates, which shows up as better prompt adherence and more sensible compositions, at the cost of a little latency. It's also the model that sits at the top of image leaderboards, which is why everyone keeps talking about it even while muttering about the censorship.

How it works. Standard pack machinery - submit a workflow to orchestration.civitai.com, long-poll, convert the output back to an IMAGE tensor. The input list is where Pro earns its name. resolution goes 1K/2K/4K - the 4K step is native to this model and costs accordingly. aspect_ratio spans 21:9 to 9:16, default 1:1. num_images (1-4) picks how many outputs you get, and there's an images socket for reference-image generation - Pro takes reference images and, combined with a good prompt, is the engine behind a lot of the character-consistency and style-transfer work you see posted.

Inputs that matter. prompt is required; this is where the thinking-mode payoff lives, so write a full scene, not tags. resolution = 4K when the output is going to a big display or print; drop to 1K for drafts and watch the Buzz meter thank you. images is the reference socket - wire one or more in when you want subject or style consistency across generations.

Install. With the whole pack, once:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Or ComfyUI Manager → Custom Nodes Manager → search Civitai Comfy Nodes → Install. Restart, then find it under Civitai/Image/google.

Auth and gotchas. Credentials resolve from a wired Civitai Auth node, then CIVITAI_API_TOKEN, then the stored key or sidebar OAuth. The pack's error hints are worth memorizing: 402 = insufficient Buzz (the one you'll hit first with Pro's 4K pricing), 401 = token missing/expired, 429 = rate limited. Thirty-minute timeout, early-preview pack, shapes can shift between releases.

The two things to know before you spend on it: Pro is a closed Google model with heavy, opaque filtering - "powerful but unimaginative" is the community's standing verdict, and Civitai's web-generator restrictions apply on top, so don't expect it to draw what the website won't. And it's the most expensive node in the image lineup by a margin. Use 4K sparingly, and Pro is the best raw-quality generator Civitai offers inside ComfyUI - just keep your expectations about what it'll agree to draw realistic.