LC MiniMax H3 Pipe Out
Unpacking
- pipe
- pipe
- fl2va_model
- fl2va_clip
- ref2va_model
- ref2va_clip
- video_vae
- audio_vae
- width
- height
- length
- frame_rate
- ref_image_0
- ref_image_1
- ref_image_2
- ref_image_3
- ref_image_4
- ref_image_5
- ref_image_6
- ref_image_7
- ref_image_8
- ref_video_0
- ref_video_1
- ref_video_2
- ref_video_audio_0
- ref_video_audio_1
- ref_video_audio_2
- ref_audio_0
- ref_audio_1
- ref_audio_2
LC MiniMax H3 Pipe Out is the mirror of LC MiniMax H3 Pipe - where its sibling bundles everything onto one wire, this one unpacks it again. Feed it an LC_H3_PIPE and out the other side come the UNETs, CLIPs, VAEs, the size fields, and every reference-media socket, ready to plug into ComfyUI's native MiniMaxH3ImageToVideo / MiniMaxH3ReferenceToVideo nodes.
Why would you bundle and immediately unbundle? Because the wire, not the value, is the point. Once you've got an H3 pipe on the canvas you can route it through a switch, tee it to two samplers, or edit it mid-flight with the Pipe node, and the Pipe Out at the end is where all those branches reconverge into explicit sockets. It's the standard context-bus pattern from the plumbing layer: bundle for legibility where the graph gets busy, unpack for explicit wiring where you're actively iterating. It's also just a great way to see what's actually inside a pipe - unconnected values come out as nothing, so a quick look tells you which slots you forgot to fill.
What comes out
The full slot list matches the Pipe node exactly: fl2va_model / fl2va_clip (for first-last-frame-to-video), ref2va_model / ref2va_clip (for reference-to-video), video_vae / audio_vae, then width, height, length, frame_rate, and the references - ref_image_0…8, ref_video_0…2, ref_video_audio_0…2, ref_audio_0…2. Plus the pipe itself passes straight through, so you can daisy-chain an unpack into more pipe plumbing without re-bundling.
Two things worth flagging:
ref_image_0=<Picture 1>. MiniMax's prompt tags are 1-based, so the image in socketref_image_0is<Picture 1>in your prompt,ref_image_1is<Picture 2>. Get this wrong and your references silently shuffle by one.width/heightcan be sourced from upstream. If an Aspect Ratio Simplifier pipe (or LC Pipe) was merged into the H3 pipe before this node, those dimensions ride along - so your output resolution can be driven by the same per-project ARS setup as everything else, without hand-typing pixels twice.
Install
Same pack, same one-time setup as LC MiniMax H3 Pipe:
cd ComfyUI/custom_nodes
git clone https://github.com/lonecatone23/ComfyUI_LC123_nodes
or install via ComfyUI Manager (search "ComfyUI_LC123_nodes") and restart. No extra pip packages - the pack runs on ComfyUI's own torch/numpy. The heavy downloads are on the model side: the H3 weights (~42.5 GB full precision) plus the video and audio VAEs, and those come with the license caveat that applies to the whole H3 stack - the MiniMax H3 Community License excludes the US, EU, UK and South Korea from running the local weights, outputs included. The plumbing is free; the thing it feeds may not be legal in your zip code.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| pipe | LC_H3_PIPE | LC MiniMax H3 Pipe. |
Outputs (29)
| Name | Type | Description |
|---|---|---|
| pipe | LC_H3_PIPE | — |
| fl2va_model | MODEL | — |
| fl2va_clip | CLIP | — |
| ref2va_model | MODEL | — |
| ref2va_clip | CLIP | — |
| video_vae | VAE | — |
| audio_vae | VAE | — |
| width | INT | — |
| height | INT | — |
| length | INT | — |
| frame_rate | INT | — |
| ref_image_0 | IMAGE | — |
| ref_image_1 | IMAGE | — |
| ref_image_2 | IMAGE | — |
| ref_image_3 | IMAGE | — |
| ref_image_4 | IMAGE | — |
| ref_image_5 | IMAGE | — |
| ref_image_6 | IMAGE | — |
| ref_image_7 | IMAGE | — |
| ref_image_8 | IMAGE | — |
| ref_video_0 | IMAGE | — |
| ref_video_1 | IMAGE | — |
| ref_video_2 | IMAGE | — |
| ref_video_audio_0 | AUDIO | — |
| ref_video_audio_1 | AUDIO | — |
| ref_video_audio_2 | AUDIO | — |
| ref_audio_0 | AUDIO | — |
| ref_audio_1 | AUDIO | — |
| ref_audio_2 | AUDIO | — |