MiniMax H3 Easy Segment Decode

Every pipeline needs its unglamorous finisher, and for the MiniMax H3 Easy context-segment chain that's Segment Decode. It takes the segments result from Segment Sample (first pass) or Segment Refine (refined) and turns it into a real, playable VIDEO - with audio muxed in. One input, one output, no settings to fiddle with. If you've got a finished segment chain, this is the node you'll click into a video preview or save node and call it a day.

Why it's built the way it is

The interesting part is how it decodes. A long multi-segment video decoded naively would mean holding the entire RGB timeline in memory at once - a one-minute 720P clip is a lot of frames, and it only gets worse with refinement. So Segment Decode works in a bounded-memory streaming fashion: it decodes one segment at a time, immediately writes those frames into a persistent FFmpeg encoder process, frees the decoded tensors, and moves on. The whole video is assembled on disk, not in VRAM. That's the same reason it exists as a separate node rather than being folded into the sampler - it's a dedicated memory strategy.

The audio side matters too, because H3 generates audio natively. When the plan has no driving audio, the node decodes each segment's audio latent through H3's audio VAE and trims it to match the delivered frames. When you used a Digital Human driving track, that one audio item is sliced across the full timeline and becomes the soundtrack instead. Either way you get a complete VIDEO with muxed audio out of the preview socket - no separate audio file to babysit.

The one hard requirement

Segment Decode needs FFmpeg. There's no way around it - the node streams raw frames into an encoder, and without FFmpeg it fails with a clear message. The pack prefers imageio-ffmpeg (which is bundled with ComfyUI itself, so most installs already have it) and falls back to an ffmpeg binary on your system PATH if that's all you have. It even respects VHS_FORCE_FFMPEG_PATH if you've pointed ComfyUI's video nodes at a specific binary. If you hit the error, install one of those two and re-queue - it's a five-minute fix, not a config labyrinth.

Output encoding is a solid default: libx264, veryfast preset, CRF 18, yuv420p, faststart for web playback. In other words, a standard MP4 that plays everywhere and previews in ComfyUI.

Where it fits

Context Segments → Segment Sample → Segment Decode   (first-pass preview)
Context Segments → Segment Sample → Segment Refine → Segment Decode   (final)

Same node, same input type, both chains feed it - you just pick which one you want to look at. It's not an output-only dead end, either: preview is a normal VIDEO socket, so it plugs into whatever video preview or save node you already use.

Installing

It's part of ComfyUI-MiniMaxH3-Easy, installed the same way as the rest of the pack. Update ComfyUI to a build with the official MiniMax H3 nodes first, then use ComfyUI Manager (search ComfyUI-MiniMaxH3-Easy, pick Nightly - the README is explicit that Nightly is the current release) or:

cd ComfyUI/custom_nodes
git clone https://github.com/nkxx188/ComfyUI-MiniMaxH3-Easy.git

Restart ComfyUI, make sure FFmpeg is available, and the workflow/MiniMax_H3_Easy_Context_Segments.json example shows the whole chain including this node in its proper place at the end.