arkennemasis fal · MiniMax Video 01 Live · $0.5/run

Here's the honest version: video-01-live takes the same two widgets as plain Video 01 - prompt and prompt_optimizer - costs the same $0.50 a clip, and fal's own one-line description of both models is literally the same sentence. "Live" is a different checkpoint on MiniMax's side, not a clickier version of the node.

I'm not going to invent a difference for you. There isn't a published spec sheet separating them in the node's fields, and this pack has essentially no Reddit footprint to check against - searching for it returns nothing, so there's no community consensus to borrow either. What I can tell you is that the difference exists at the model level and is free to test, because the node is a drop-in swap.

Test it like a grown-up instead of guessing

The stupid way: duplicate the branch, run it, and pay for both clips. The pack has a better way. Shot Selector (in arkennemasis/Utility) runs only the first N branches in order, or a random sample from a seed - and the unselected branches never execute, so a paid fal node downstream is never called. Put Live and base Video 01 behind a selector, run N=1, compare the two takes across a couple of prompts, and you'll have spent a dollar instead of ten finding out which one suits your subject.

That's the general pattern for this whole pack, by the way. Lazy branch gating is the reason a canvas of thirty paid nodes doesn't mean thirty charges.

How the run works

Same pipeline as every fal node here: the node reads FAL_KEY from the .env in your ComfyUI install, validates the inputs, estimates the cost from what's connected, and stops before uploading if the estimate is over max_cost_usd (default $20; 0 disables the cap). Then one submit - the only billable step - a poll, and a download into output/fal/fal-ai_minimax_video-01-live/. The node shows queue position and elapsed seconds; ComfyUI's Cancel cancels the fal request as well. Everything is logged to output/fal/_requests.jsonl.

max_concurrent (default 1, up to 32) is shared across every fal node in the run: at 1 they go one at a time, and if one fails, the ones still waiting are never started. reuse_identical_run (default true) hands back a clip you already paid for when the inputs are identical and the file is still on disk - handy for these A/B experiments, but remember to switch it off when you actually want a second sample.

Inputs and outputs

prompt and prompt_optimizer are required. prompt_optimizer on (the default) means the model rewrites your prompt before generating - the usual suspect when your words seem to have been ignored.

Outputs: video (a ComfyUI VIDEO value - save it, dub it, lip-sync it, assemble it), video_path (STRING path under output/fal/fal-ai_minimax_video-01-live/), and info (JSON with the request id, cost estimate, files and fal's answer).

No seed, no duration, no resolution. Expect no usable audio track; MiniMax's native audio work is the H3 line.

Install

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

Restart, or install through ComfyUI-Manager (search arkennemasis). requirements.txt is httpx, and that belongs to the pack's Codex nodes - the fal provider needs nothing beyond the standard library. Then:

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

next to run_nvidia_gpu.bat.

Where people get burned

• You expected a visible difference in the node. There isn't one; the difference is the model behind it. Run the A/B.
• You paid twice for the same prompt. That's reuse_identical_run off. It's on by default for a reason.
• Nothing was charged but nothing happened. Fine - that's the pre-flight refusing on max_cost_usd, or an empty prompt. Failures before the submit are free by design.
• ComfyUI restarted while it was running. The job continued on fal and was billed. Grab the id from output/fal/_requests.jsonl and pull it down with fal Recover Result - that one's free.