Flux Ultra (fal)

The Flux you run locally is Flux.1 [dev] - the open-weights model. FluxUltra is the other one: Black Forest Labs' closed, hosted Flux 1.1 [pro] Ultra, the top of their paid ladder, reaching it through fal instead of a checkpoint on your drive. Nothing downloads, nothing touches your VRAM. You type a prompt, fal runs it on BFL's endpoint, and an image comes back. You reach for this when you want the pro-tier look - the higher-resolution, cleaner output that the open dev weights don't quite hit - and you'd rather pay per image than not have access at all.

It's part of gokayfem's ComfyUI-fal-API pack (same author as the well-known VLM Nodes), which is basically a giant set of thin wrappers around fal's model catalog. One FAL_KEY and you can call any of them.

How it works

Flux is Black Forest Labs' transformer-based text-to-image family (~12B params, dual CLIP-L + T5-XXL text encoders). The open [dev] weights are distilled and guidance-based; the Pro and Pro Ultra tiers are the undistilled, higher-quality models BFL keeps behind their API. This node is a passthrough to that API: your inputs get packaged into a fal request, the result URL is downloaded, and you get a native ComfyUI IMAGE out the other end.

The inputs and outputs that matter

Only a handful actually move the needle:

• prompt - Flux reads natural language well, so write sentences, not a tag soup.
• aspect_ratio - this is how you set shape, not width/height. Options run 21:9 through 9:21, default 16:9.
• raw (default off) - the one worth knowing about. Ultra's "raw" mode dials back the polished, over-produced AI look toward something more candid and photographic. Flip it on when your outputs feel too glossy.

The rest: safety_tolerance is a 1–6 scale where 1 is strictest and 6 is most permissive; enable_safety_checker toggles the post-gen filter; num_images is capped at 1 on this node; sync_mode controls whether the call blocks; and seed (optional, -1 for random) lets you reproduce a result. Output is a single IMAGE - wire it straight into a Preview or Save node.

How to install it

Via ComfyUI Manager: open the 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

Then restart ComfyUI. There are no model files to download - that's the whole point of an API pack. The only dependency that matters is fal-client, which requirements.txt pulls in.

Before anything runs you need a key. Grab one from fal.ai, then either copy config.ini.example to config.ini inside the pack folder and paste your key under [API], or set export FAL_KEY=your_key. The node lives under the FAL category once ComfyUI restarts.

Common issues & troubleshooting

• Auth errors / "invalid key." No FAL_KEY, or a typo'd one. Set it, restart, and remember config.ini is gitignored so it won't leak into a commit.
• It costs money and you can feel it. Every run is a real fal API call billed to your account - Ultra is one of the pricier endpoints. The pack ships a persistent result cache (an identical prompt+settings call is served free from disk; use force_rerun to bypass), a spend guard, and a cost estimator precisely because this bills per image. Turn the spend guard on if you're experimenting.
• A run fails with a wall of text. Good - the pack surfaces fal's actual error instead of silently returning a blank. It'll usually name the problem: a content-policy rejection, a quota issue, or a bad parameter. Read it before you guess.
• Windows Portable can't find fal-client. Install it into the embedded Python: python_embeded\python.exe -m pip install fal-client.

One honest note: if your goal is local, free, and unlimited, this isn't it - that's Flux dev. FluxUltra is for when you specifically want BFL's paid top tier and the API is the only door to it.