MiniMax H3 Aicg 宽高比

MiniMax H3 Aicg 宽高比 has exactly one input, one output, and no settings. It is also the node that saves you from the dumbest recurring mistake in vertical-video work: you set 9:16 on your H3 node, then something downstream - a save node's resolution preset, an upscaler, a compositor - is still sitting on its own 16:9 default, and you find out when the file comes out letterboxed.

What it is

A pass-through for one value: the aspect ratio you resolved on the H3 side. The description is as literal as it gets - "keep downstream resolution selectors aligned with MiniMax H3 Aicg without copying width or height." You feed it an h3_context; it hands back an aspect_ratio value. No width, no height, no maths, no magic.

Why a whole node for that

Because H3 nodes don't expose aspect ratio as a wire. It's a dropdown widget on the node, and in ComfyUI a widget value is not connectable - you can't drag 9:16 out of it. So every other node in your graph that has its own aspect dropdown gets re-picked by hand every time you change your mind about the format. Do that three times across a six-node graph and you will eventually miss one.

The context bus pattern is what makes a one-line fix possible: the H3 nodes already stuff the resolved aspect ratio into the h3_context bundle, so this node is a three-line extractor sitting on top. It looks up the label for the context's aspect ratio and returns it, and it throws a clear error if you handed it something that isn't an H3 context. There is no per-frame cost and nothing to configure.

Inputs and outputs that matter

All of them: h3_context in (from MiniMax H3 Aicg, 上下文分段, or anything else that emits a MINIMAX_H3_CONTEXT), aspect_ratio out.

One thing to internalise: the output is the label string, not numbers. 9:16, 16:9, 21:9, and the rest of the ladder - 1:1, 2:3, 3:2, 3:4, 4:3. It's typed as a wildcard precisely so it can be dropped into any downstream combo widget that speaks the same vocabulary. Which is why it works beautifully with a save node whose ratio dropdown uses the same labels, and does nothing useful for a node that wants pixel integers.

If you need pixels, you don't want this node - read the latent dimensions from MiniMax H3 Aicg 输出, or wire width/height explicitly.

Where it sits in a workflow

Typical use: a vertical short-form pipeline. H3 generates at 9:16, AICG-渲染器（一采） gives you a VIDEO, and somewhere later a save/encode node wants to know the shape so it doesn't pad or crop. You put this node off to the side, tap h3_context from the H3 node, and wire its aspect_ratio into that downstream selector. Now the format is decided in exactly one place.

It also earns its keep in the pack's own segment workflows, where the same ratio has to hold across a whole chain of segments - the last thing you want is segment four deciding it's widescreen.

Installing

Nothing special: it ships in comfyui-AICG3D. Manager → search AICG3D / 短剧工作台, or:

cd ComfyUI/custom_nodes
git clone https://github.com/JGRFW/comfyui-AICG3D

Restart, right-click → AICG3D.

Where people get burned

Two ways. First, assuming it outputs a resolution - it doesn't, and a node expecting two integers will refuse it. Second, the stale-context trap that comes with any bundle-based pack: if you insert this node before the node that decides the aspect ratio (say, upstream of a context-segments node that re-resolves it), you're extracting the ratio from an earlier version of the plan. Put it downstream in the wire sense, or at least off the same context you're actually rendering.

And if you swap between an AICG3D_H3 node and a 上下文分段 node, re-check that the downstream selector still matches its label set. The error message if it doesn't ("Unsupported MiniMax H3 aspect ratio") is accurate but it won't tell you which node to fix.