arkennemasis fal · MiniMax H3 Max Styles Retro Toon 70s · $0.08/s

The 1970s animated-feature look, baked in

Retro 1970s hand-painted animation: think the era of cel paint, soft outlines, and backgrounds that were clearly made by a person with a brush. This endpoint generates 768p video with audio in exactly that style from a text prompt or an optional first frame, and it applies the look automatically.

Why would you want an endpoint for a style instead of a prompt? Because style prompts are where video models are least reliable. You can write "1970s hand-painted animation, cel shaded, gouache backgrounds" and get something that nails it once and wanders off on the next shot. Locking the look into the endpoint means your only remaining variable is the content - which is what you want when you're producing a sequence and not a single happy accident.

$0.08 per second at 768p, audio included. Five seconds runs $0.40; the fifteen-second maximum $1.20.

Inputs that matter

• prompt - scene, action, camera, sound. Don't describe the style; describe what happens in the frame.
• image - optional first frame, and the tooltip is pointed about it: use painted-style artwork for the most consistent look. If you have a concept painting in the right idiom, this is the highest-leverage input on the node. If you feed it a photo, you're asking the endpoint to translate, and translation always shows.
• duration - seconds, default 5, 5–15 supported.
• aspect_ratio - 21:9, 16:9 (default), 4:3, 1:1, 3:4, 9:16, and ignored when an image is supplied.
• seed - -1 by default; fix it to repeat a take you liked.

All the fal-node guards from this pack come along: max_cost_usd ($20 default, stops the run before upload if the estimate exceeds it, 0 disables), reuse_identical_run (returns the paid file for identical inputs instead of buying it twice), extra_json and max_concurrent (default 1 paid call at a time across the run).

Outputs

video (to SaveVideo, PreviewVideo, or the pack's Video Assemble), video_path, seed and info (request ID, estimate, saved files, fal's answer verbatim).

Install

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

Restart ComfyUI, then find it under arkennemasis/fal/Video/MiniMax. Key in the .env next to run_nvidia_gpu.bat:

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

Where people get burned

Double-describing the style. The endpoint paints 70s cel animation whether or not you ask. Style keywords in the prompt add noise and can drag the result toward a modern imitation of the era rather than the era itself.

Hands and text. Hand-painted animation had a visual grammar, but generative models still fumble fine anatomy and lettering. Close-ups on hands remain a coin flip; keep them out of the shot list if you're on a budget.

Assuming the audio will be period-correct. You get generated audio, not a vintage sound design pass. Dialogue and ambience come out sounding like the model's own idea of the scene - which is often fine, occasionally jarring against the visuals.

Long clips and story beats. Fifteen seconds is the ceiling and rarely the right length. Build a sequence of 5-second shots and cut them; the pack ships the assembly nodes (Video Assemble, Caption Style, Load Clips) for exactly that job.

The exit door is closed in both directions. Hosted generation means prompts and first frames leave your machine, and there's no local equivalent of this specific endpoint's look. Decide on the project, not on the fly.