RunningHub MiniMax H3 T2VA Target (Legacy)

Before the pack got its one-box generation nodes, a MiniMax H3 workflow was a chain: load model, load text encoder, load VAEs, then a target node to say how big and how long the output should be, a text-encode node, an encode node, a sampler, a decoder. RHMiniMaxH3T2VATarget is the "how big and how long" node for the text-to-video path, and it's a great example of a legacy node that's only interesting if you inherited an old workflow.

What a "target" actually is

In H3 terms, a target is the resolved output spec: width, height, frame count, fps, and - this is the part that doesn't exist in normal image workflows - the latent shapes for the video stream and the audio stream separately. The target node is what converts your loose "16:9, five seconds" request into the exact numbers every other node in the chain trusts. That's why it returns a shape_info string alongside the target handle: a JSON dump of those resolved values, handy for debugging why something came out 832×480 instead of 1280×720.

Inputs

• aspect_ratio - auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16.
• duration_seconds - must be explicit, 4.0–15.0. There's no auto here; unlike the reference path, text-only generation has nothing to infer a length from, and the node validates the range at execution time.
• width / height - optional, both default 0, which means "compute from aspect ratio." If you fill them, fill both, and keep them multiples of 32 - the validation enforces it.

Outputs

target (type MINIMAX_H3_TARGET) and shape_info (STRING). The target wires into the encode node in the old chain.

Should you use it?

No - not for anything new. It's deprecated and hidden from search, kept alive so old serialized workflows still instantiate and run. RHMiniMaxH3VideoGen builds its T2VA target internally from the same aspect_ratio / duration_seconds / width / height widgets, with zero extra nodes. The only reasons to touch this node are: you have a pre-refactor workflow you're nursing along, or you're deliberately building the manual legacy chain (which the current generation nodes have fully absorbed). If you're mid-migration, the pack ships a workflow migrator - python3 tools/migrate_workflow.py old_workflow.json --in-place - that rewrites old node IDs and widget layouts for you.

Install

Same pack-wide routine:

cd ComfyUI/custom_nodes
git clone https://github.com/RH-RunningHub/ComfyUI-RH-MiniMax-H3.git
pip install -r ComfyUI-RH-MiniMax-H3/requirements.txt

Restart, and have the ~95 GiB INT8 ConvRot bundle in models/MiniMax-H3-INT8-CONVROT/ (see the pack README for the hf download command). Honestly, the fastest way to understand this node is to open an old workflow that uses it, run the migrator, and watch it disappear into RHMiniMaxH3VideoGen - that's the intended retirement plan.