🎨 Nano Banana Pro (Gemini 3 Pro)

This is the flagship end of Google's Nano Banana line - Nano Banana Pro, aka Gemini 3 Pro Image - as a ComfyUI node. If the plain Gemini Flash 2 node in this pack is the quick-and-cheap tier, this is the one you reach for when you actually care about the result: native 4K output, up to 14 reference images, and a model that reasons about your prompt before it draws ("thinking mode"). It's an API call, not a local model, so it needs a Google key and it doesn't touch your GPU.

The reason it's a distinct node and not just a dropdown is that Pro exposes controls Flash doesn't. You get real aspect-ratio and resolution presets instead of hoping the model guesses, and it returns text alongside the image - so you can see what the model "thought" it was making.

How it works

Your prompt and any reference images go to the Gemini 3 Pro Image API; an image and a text blurb come back. The model is genuinely strong at multi-language text rendering and at holding a subject consistent across references, which is what the up-to-14-images support is for. Feed it a batch of reference shots through the single image input and it can compose or keep a character consistent across them. As with the rest of this pack, the text fields support Google Translate and dynamic prompts, so wildcard syntax and non-English prompts both work.

The inputs and outputs that matter

• Gemini_API - your key, pasted here or read from the pack's API_key.json if you leave it blank.
• prompt / sub_prompt - the main instruction and a second box for extra detail.
• image (optional) - reference image(s). The model accepts up to 14; pass a batch here for multi-reference composition or character consistency.
• aspect_ratio - Auto plus the usual set (1:1, 2:3, 3:2, 16:9, 9:16, 21:9, and more). Pick one rather than trusting Auto if framing matters.
• resolution - 1K, 2K, or 4K. This is the cost dial: 4K is the expensive setting, so don't leave it there for drafts.

Three outputs: image, text_output (the model's accompanying text), and prompt (the final prompt actually sent, after translation).

How to install it

In ComfyUI Manager: search SDVN_Comfy_node, install, restart. Or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root and restart. Get a Gemini key at aistudio.google.com/app/apikey, and prefer putting it in API_key.json (rename API_key.json.example) over pasting it into the node, so it doesn't ride along in your saved workflow.

Common issues & troubleshooting

It's not free, and 4K adds up. Pro pricing runs roughly $0.04 to $0.24 per image depending on resolution. If your bill surprises you, it's the resolution and reference-image count - both push token cost up. Draft at 1K, finalize at 4K.

Refusals and missing celebrities. Google's safety filter is strict and got stricter in January 2026, with explicit restrictions on famous people and IP. A refused prompt comes back empty; that's server-side and you can't override it from the node.

Invisible watermark. Every Gemini image ships with a SynthID watermark baked in. Fine for most uses, relevant if provenance matters to you.

Empty result with a valid key. Usually a billing/quota issue on the Google account, or a prompt that tripped the filter. Verify the key in Google AI Studio to rule the key out first.