arkennemasis fal · MiniMax H3 Max Styles VHS · $0.08/s

Analogue-looking footage, with the damage as a setting

Most VHS-look attempts are a post-process: generate something clean, then smear a grain and a chroma-shift over it. This endpoint generates the tape look as part of the generation - 768p video and audio with VHS styling applied automatically, and a damage_level dropdown that decides how far gone the tape is.

That dropdown is the reason it's worth a dedicated node. light gives subtle analog noise, medium (the default) gives visible damage, and heavy adds strong tracking errors and distortion. You pick the era and the wear in one go instead of guessing at a grain node and hoping.

Where it's genuinely useful: music videos, "found footage" horror, anything set in a decade with a camcorder in it, and that perpetual request for a clip that looks like it was recorded off broadcast. What it's not is a clean-up tool - there's no de-VHS direction here.

$0.08 per second at 768p, audio included. Five seconds $0.40, fifteen $1.20.

Inputs that matter

• prompt - scene, action, camera and sound. The style and the tape damage are applied for you.
• damage_level - light, medium (default), heavy. Start at light and step up: heavy is a look, but it eats a lot of the picture, and it's harder to judge whether the underlying generation was any good.
• image - optional first frame. The tooltip's tip is specific to this style: use a 4:3 image for the classic VHS frame. That's the aspect the format actually was, and it does more for believability than any amount of added noise.
• duration - seconds, default 5 (5–15 supported).
• aspect_ratio - 21:9, 16:9 (default), 4:3, 1:1, 3:4, 9:16; ignored when an image is supplied.
• seed - -1 by default, and worth setting, because with this much texture applied you'll want to reproduce a specific take.

Plus the pack's guards on every fal node: max_cost_usd (default $20, evaluated before upload so a run over your cap costs nothing), reuse_identical_run (identical inputs come back from disk instead of fal), extra_json, max_concurrent (default 1 paid fal call at a time across the graph).

Outputs

video (VIDEO - into SaveVideo, PreviewVideo, or Video Assemble), video_path, seed and info with the request id, the estimated cost, the saved files and fal's raw response.

Install

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

Restart ComfyUI, then it's under arkennemasis/fal/Video/MiniMax with the other style nodes. Key line in the .env beside run_nvidia_gpu.bat:

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

Where people get burned

Heavy damage hides a bad generation. If the motion was weak, heavy makes it hard to tell - for you and for anyone giving feedback. Judge the clip at light first.

16:9 default fights the aesthetic. The default aspect ratio is widescreen because the endpoint's default is; VHS wasn't. Choose 4:3, or feed a 4:3 frame, if the point is the format.

Per-second pricing makes experimentation linear. Every attempt at a 15-second clip is triple a 5-second one. Do the look test short.

There's no undo. This endpoint only goes one way - tape-styled generation. If you wanted the same shot clean as well, run the plain H3 Max text-to-video node alongside it and keep both.

Stacks don't compose the way you'd hope. Adding a film-grain or chromatic-aberration node after generation can be nice, but the endpoint has already done the era's colour response; over-processing this look goes to "broken video file" fast.