arkennemasis fal · MiniMax Image 01 · $0.01/image

One cent an image. A dollar buys you a hundred. That's the entire pitch, and it's a good one - not because MiniMax Image 01 beats your local checkpoint on quality, but because there are jobs where ten cents of throwaway images is exactly what you want and firing up a 12GB model to get them is silly.

What it's for

Think of it as a cheap b-roll and concept machine. Generating forty thumbnail comps, mocked-up backgrounds for a product pipeline, or a wall of stills to choose between. Fal's own note on the model says longer prompts produce better images, which is worth remembering because short tidy prompts are the instinct for a cheap model and the wrong one here.

Where it doesn't belong: the hero image. You have Flux, SDXL, Z-Image, whatever your checkpoint of choice is, running on your own GPU where it's free and unbounded. Use those. Use this when the answer you want is "twenty options, fast, and I don't care if sixteen of them are bad."

How it works

It's the pack's thin fal wrapper around fal-ai/minimax/image-01. The node reads FAL_KEY from the .env beside your launcher, checks the inputs (including a 1500-character ceiling on the prompt), estimates the cost against the cap, submits one billed request, polls, and downloads results into output/fal/fal-ai_minimax_image-01/. There's no model file, no VRAM budget, no local install - MiniMax is closed and API-first, so hosted is the only door, and at a penny a call the pricing is the feature.

The inputs that matter

• prompt - up to 1500 characters. Go long. This isn't a model that rewards haiku.
• aspect_ratio - 1:1, 16:9, 4:3, 3:2, 2:3, 3:4, 9:16, 21:9. Pick the one your delivery needs now rather than cropping later.
• num_images - 1 to 9. Nine images in one call for 9 cents, and the node's price estimate follows this number.
• prompt_optimizer - default off here, unlike the video nodes where it's on. Flip it on when you'd rather the model expand your prompt than take it literally.

Optional controls are the four pack-wide ones: max_cost_usd (default $20; the run stops before uploading when the estimate is over it, 0 disables), max_concurrent (default 1 - one paid fal call at a time across a run), reuse_identical_run (default on; an identical paid-for run returns the saved files instead of billing again, restart included), and extra_json for fal fields the node doesn't expose.

Outputs: images - all of the returned pictures come back concatenated onto one batched IMAGE output, so you can send them straight into a Save or the pack's Option Board to compare a run at a glance; image_paths, the newline-joined file paths; and info, the JSON with request id, estimate and fal's answer.

Install

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

Or ComfyUI Manager → Install via Git URL. Restart, then put the key where the node actually looks:

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

in the .env next to run_nvidia_gpu.bat (or in the ComfyUI folder). It lands under arkennemasis/fal/Image/MiniMax.

Honest notes

num_images at 9 is the move for a cheap selection pass, but be aware the node warns if fal returns pictures at mismatched sizes: the IMAGE output carries only those matching the first, though every file is still saved to disk. With a single call that shouldn't happen; if you see the warning, your files are fine, the batch is just trimmed.

And the standing caveat for any API node: your prompt leaves the machine, and this pack - like every pack in this category - holds a credential and talks to the network by design. Prefer packs you can read; keep your fal key scoped to fal and nothing else. A penny an image is a bargain; a leaked key is not.