Flux Pro 1.1 (fal)

This is the Flux Pro you actually want to start with. Flux Pro 1.1 is BFL's refreshed flagship - faster and generally cleaner than the original Pro - and, like everything in the Pro tier, it lives only on the API. BFL never released Pro weights, so FluxPro11_fal calling fal's flux-pro/v1.1 endpoint is the way you run it from ComfyUI. It renders on fal's hardware and drops a native IMAGE into your graph.

Because Pro is a managed, opinionated endpoint, this node is deliberately simple. You don't get a steps slider or a guidance dial - BFL tuned those in-house - so you point it at a prompt and a size and let it cook. That's a feature if you just want good Flux output without fiddling.

The inputs that matter

• prompt - natural language. Flux's T5 encoder parses full sentences; write like you're describing the shot to a person.
• image_size / width / height - pick a preset, or go custom (width/height cap at 1440, steps of 32).
• num_images - batch up to 10 in one call.

The rest: safety_tolerance is a 1-to-6 dial (higher = more permissive), seed is -1 for random, and sync_mode controls whether the call blocks until the result is ready. There's a single IMAGE output - wire it to Preview or Save Image.

How it works

No model touches your machine. The node hands your parameters to fal, fal runs Pro 1.1, and the finished image comes back over the wire as a ComfyUI image. The pack keeps a persistent result cache, so an identical prompt-and-settings combo is served free and instant on the second call; use force_rerun when you want a genuinely new roll.

Installing it

ComfyUI Manager → search ComfyUI-fal-API → install → restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt

There are no weights to download - an API pack has nothing to cache locally. Set your fal key once (copy config.ini.example to config.ini and paste it, or export FAL_KEY=your_key; get the key at fal.ai/dashboard/keys), restart, and find the node under FAL.

Where people get burned

The missing steps/guidance controls trip people who expect a full local-style Flux graph. They're gone on purpose - Pro is a hosted product, not a checkpoint you configure. If you want that level of control, use FluxDev_fal or FluxGeneral_fal instead, which expose the knobs.

And the money is real: each generation is a fal charge, so the result cache is your friend during iteration. safety_tolerance loosens the filter within limits but doesn't disable it - refused prompts come back as visible fal errors that name the reason. If you want editing rather than text-to-image, the Kontext nodes (FluxProKontext_fal) are the ones that take an input image.