arkennemasis fal · Flux Pro v1.1 Ultra · $0.06/image

Flux Pro's whole thing is that you can't have it. The [pro] tier never shipped weights - it lives behind Black Forest Labs' API and the resellers, and this node is the fal.ai door to its 1.1 Ultra revision. You get the undistilled base model, up to roughly 2K, for six cents an image, and your GPU sits there doing nothing.

That sounds dismissive. It isn't. Pro is also the tier that doesn't have Dev's same-face problem: the distillation that made Flux Dev and Schnell fit on consumer cards is what flattened the facial variety, and Pro skipped all of it. If you want one hero image at 2K and you don't want to download twelve gigabytes to get it, this is the cheapest honest route.

What's actually happening when you hit Run

This node is one of the fal family in comfyui-arkennemasis - a pack of ~106 image nodes and 62 video nodes where each fal model gets its own node, generated from fal's own schema, all sharing the same run machinery. Nothing local samples anything. The prompt and any reference image you connect get checked, priced, uploaded to fal's CDN, submitted as one billed job, polled until done, and the result is downloaded into output/fal/fal-ai_flux-pro_v1.1-ultra/.

Because the code never touches a checkpoint, the node is a thin client - and I mean that as a compliment: the fal side of this pack uses Python's standard library only, so installing it adds no dependencies to your ComfyUI. (The httpx in requirements.txt is for the pack's separate ChatGPT/Codex nodes, not this one.)

The inputs a beginner actually sets

• prompt - full sentences. The T5 encoder behind Flux reads natural language; tag soup underuses it.
• aspect_ratio - defaults to 16:9. Billing is per image here, not per megapixel, so a big output costs the same as a small one. Pick the shape you actually want.
• raw - off by default. Turning it on gives you less processed, more natural-looking output. This is the switch people leave alone for a month and then wonder how they missed.
• enhance_prompt - lets the model rewrite your prompt. Handy when you're stuck, completely unwanted when you've tuned a prompt you like.
• safety_tolerance - 1 strictest, 5 most permissive. 2 is the default.
• num_images - up to 4. Each one is another six cents.
• image + image_prompt_strength - wire a picture in and it acts as an image prompt, not a full img2img. Strength defaults to a very timid 0.1; only relevant if you connect something.

There's also seed (use a real number once you get a keeper, so you can hold it while you edit the prompt) and the shared money widgets: max_cost_usd, reuse_identical_run, extra_json, max_concurrent.

What it hands back

images is the IMAGE socket - wire it into SaveImage, an upscaler, a video node, whatever. image_paths gives you the saved file path as a string. seed comes back as an actual integer so you can pin it, plus prompt, timings, has_nsfw_concepts, and info, which is JSON containing the fal request id, the cost estimate, the files written and fal's full response. On the photorealism side: Flux is photorealistic out of the box, and stacking "ultra realistic, 8K, masterpiece" on top of it mostly makes things look over-processed. Spend your tokens on lens, light and film stock instead.

Install

In ComfyUI Manager search for comfyui-arkennemasis (the pack title) and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis

Restart ComfyUI. Then put your key in the .env file of the ComfyUI install - for the portable build, the folder holding run_nvidia_gpu.bat:

FAL_KEY=your-key-from-fal.ai/dashboard/keys

There is no key box on the node, deliberately: a typed key gets stored in every workflow you share. The cost guard is max_cost_usd, default $20, and it stops the run before anything is uploaded.

Where it bites

The obvious one first: no FAL_KEY, no run - the node tells you it's missing rather than failing cryptically, and nothing is billed. Second, a fal account with an exhausted balance simply stops generating; that's the platform, not the node. Third, reuse_identical_run is on by default, which means repeating the exact same run returns the cached file instead of charging you again - great until you want a fresh take from the same settings, at which point you turn it off. And if ComfyUI dies mid-job, your money isn't gone: every request is logged to output/fal/_requests.jsonl, and the pack's free Recover Result node collects a finished job by id without paying twice.

One caveat worth stating plainly: this pack has essentially no Reddit footprint I could find, so there's no crowd telling you it's great. Read the source - it's short, MIT, and the money handling is unusually careful for a wrapper this size.