MiniMax H3 Video Settings

MiniMax H3 has two unforgiving grids you have to hit: the canvas is rounded to multiples of 32 pixels, and the frame count has to land on the model's "17k+5" rhythm - 124 frames is ~5 seconds, 362 is ~15, and a number off-grid gets snapped anyway, so the widget you set isn't always the value you get. Video Settings is the node that does that arithmetic for you in one place, so you stop retyping 1344 x 768 and a hand-computed frame count into every node in the workflow.

The pain it removes is the same repetition the whole plumbing layer of ComfyUI fights: one value wired to five consumers, each retyped, each silently capable of drifting out of sync. This node is the "one source" end of that fight - pick an aspect ratio and a megapixel budget, get width and height; type a duration in seconds, get length on the correct frame grid.

The inputs and outputs that matter

• aspect_ratio - one of the eight ratios ComfyUI's Resolution Selector offers (default 16:9 (Widescreen)).
• megapixels - the canvas pixel budget, default 1.0. Note that's "1 MP = 1024×1024," not a literal million: the stock 1344×768 canvas is about 0.98 MP. The result is rounded to the 32-pixel grid.
• duration - clip length in seconds at 24 fps, default 5.0, which snaps up to 124 frames.
• advertise - the interesting one, off by default. When it's on, every MiniMax H3 Hybrid to Video node in the same graph takes width, height, and length from this node automatically: their size and length widgets grey out at once, refuse links, and any hybrid node you add afterwards arrives greyed. No wires at all.
• Outputs: width, height, length - all integers, width/height multiples of 32, length on the 17k+5 grid. The outputs are always live whether or not advertising is on.

Wiring it

Video Settings ─ width ─▶ (width)  Hybrid to Video
              ├ height ─▶ (height) Hybrid to Video
              └ length ─▶ (length) Hybrid to Video

Explicit links work everywhere, including inside subgraphs. Advertising is the no-link convenience for the root graph only - inside a subgraph you wire the outputs by hand. Only one settings node per graph advertises at a time; flipping a second one on steals the slot (the first gets switched off with a warning), and a pasted or loaded workflow that arrives with several keeps the first in node order.

Installing it

Part of ComfyUI-Arisu-Nodes (swqa7697, GPL-3.0, released September 2026), a zero-dependency pack - nothing to pip install. It does need ComfyUI ≥ 0.30.0, the build that shipped MiniMax H3 support (developed against 0.34.5). Manager: Install Custom Nodes → "ComfyUI-Arisu-Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/swqa7697/ComfyUI-Arisu-Nodes.git

Restart, then Add Node → Arisu Nodes.

The model caveat nobody should skip

Before you go far down the H3 path, know the model's context. MiniMax H3 is MiniMax's open-weight omni-modal video model, and it generates video with native audio rather than bolting on a sound pass. But its Community License excludes the EU, UK, South Korea, and the United States from the "Applicable Territory" - if you're in one of those regions, you're not licensed to run the local weights at all (the hosted Hailuo API is the licensed path there). These nodes don't change any of that: they wrap the stock H3 checkpoint/CLIP/video VAE/audio VAE you already loaded. Where the license allows, though, this settings node is the cleanest way to keep every consumer of those numbers honest.