Flux Pro (fal)

Flux comes in three tiers, and Pro is the one you can't download. Schnell and Dev shipped as open weights; Pro never left the API. So if you want Pro-grade Flux output, an API node like this one is the only door in - FluxPro_fal calls fal's Flux Pro endpoint and returns a native IMAGE you can drop straight into a ComfyUI graph. There's no "run it locally" alternative to compare against, because BFL never released these weights.

Pro sits above Dev on quality and prompt adherence, and it's the tier BFL positions for production work. The catch worth knowing up front: this is the original Pro. Its successor, FluxPro11_fal (Flux Pro 1.1), is faster and generally a better default. Reach for this node when you specifically want the original Pro's look; otherwise start with 1.1.

How it works

Everything happens on fal's side. The node uploads your prompt and settings, fal renders, you get the image back - no transformer, no T5, no VAE on your machine. Because it's a managed endpoint, you get real generation controls here that the newer 1.1 node hides: num_inference_steps and guidance_scale are both exposed, so you can tune the render the way you would on a local Flux graph.

The inputs that matter

• prompt - plain-language sentences; Flux's T5 encoder is built for them.
• guidance_scale - defaults to 3.5. As with the rest of Flux, that default leans toward the "same face" look on people; try lower for portraits.
• num_inference_steps - 28 is the standard starting point.

Also on the node: image_size presets with width/height for custom dims (capped at 1440, in steps of 32), num_images up to 10, seed (-1 = random), and safety_tolerance - a 1-to-6 dial where higher is more permissive. Output is a single IMAGE.

Installing it

Through ComfyUI Manager: search ComfyUI-fal-API, install, restart. Or manually:

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

No model files to fetch. Set your key once - copy config.ini.example to config.ini and paste the key from fal.ai/dashboard/keys, or export FAL_KEY=your_key - then restart. The node appears under FAL.

Where people get burned

The honest first note: for most people, 1.1 is the one you want. It's the newer Pro, quicker and usually cleaner, and it's a sibling node in the same pack. Don't default to this node just because it's named "Pro."

Beyond that, the standard API caveats apply. Every render is a paid fal call, so lean on the pack's persistent result cache while iterating - identical settings won't re-bill. And safety_tolerance isn't a magic bypass; the endpoint still enforces content policy, and a refused request surfaces as a readable fal error, not a blank image. If something fails, the message usually tells you exactly which parameter or policy to fix.