Nano Banana Pro (fal.ai)

Nano Banana Pro is Google's flagship image model - Gemini 3 Pro Image, native 4K output, multi-language text rendering, and a per-image price tag to match. FAL_NanoBananaPro is the text-to-image wrapper that puts it in your ComfyUI graph via fal's fal-ai/nano-banana-pro endpoint. There are no weights to download, because there are no weights. This is a rental, and it's the priciest one in this pack.

The community's honest take after a few months with it: powerful but unimaginative. Gorgeous 4K renders, excellent realism, and a safety filter plus invisible SynthID watermark on everything it makes. If you want the premium Google output without leaving your workflow, this is the node. If you want range, you'll be back on open models by dinnertime.

How it works

Prompt goes up to fal, model renders, result comes back as a batched IMAGE tensor. Same plumbing as every node in the pack - fal.subscribe behind a clean wrapper, one output, nothing local.

The inputs that matter:

• prompt (required). This is the whole game; Pro models reward a clear, detailed description.
• resolution - 1K, 2K, or 4K; default 1K. 4K is the model's headline capability and the single biggest cost multiplier. Don't run 4K for every test.
• aspect_ratio - the standard set (1:1, 4:3, 16:9, 3:2, etc.), default 1:1. Note this is not the extended ultrawide list the Nano Banana 2 node offers.
• safety_tolerance (1–6, default 4) - how much the filter lets through. It's a nudge, not an off-switch.
• enable_web_search - lets the model ground on live web results; off by default, weird and occasionally useful.
• num_images (1–4), seed (-1 random), output_format (png/jpeg/webp), private_mode.

One IMAGE output, into a Preview or Save. Notably absent here: no thinking_level knob - the Pro model's reasoning isn't exposed the way the Flash node exposes it, so what you configure is what you get.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/elle-trudgett/fal-ai-comfyui-nodes
pip install -r fal-ai-comfyui-nodes/requirements.txt

or ComfyUI Manager → ComfyUI fal.ai Nodes. No model downloads. You need a fal key: export FAL_KEY="...", or a config.ini with [API] / FAL_KEY in the node folder for ComfyUI Desktop; get one at fal.ai/dashboard/keys and restart.

Gotchas

• "FAL_KEY not set" - the pack's standard first-run error; env var or config.ini fixes it.
• Watch the bill. Pro is the premium tier - think $0.04 to $0.24 per image depending on resolution. A 4K batch of four is a real charge. Start at 1K, iterate on the prompt, promote to 4K only for the final.
• The filter and the watermark are baked in. No setting in this node bypasses Google's safety stack, and SynthID travels with the output.
• Errors come back unwrapped as fal.ai error: ....

If you already have Nano Banana 2 and it's fast enough, Pro is only worth the upgrade for 4K-native output or when the Flash tier's prompt adherence lets you down. For everything else, the open-weight models in your own ComfyUI install are free and unfiltered - this node is for when Google's ceiling is the point.