Nodes/H3 Continuum/[Legacy] H3 Continuum Finish
ComfyUI Node

[Legacy] H3 Continuum Finish

The node that trims the overlap and hands the baton to the next clip

By ukr8b3g-cmyk·Created 29 days ago·Updated a day ago· 73
[Legacy] H3 Continuum Finish
  • samples
  • images
  • audio
  • plan
  • images
  • audio
  • state
  • report

H3 Continuum Finish is the other half of the pack's V1 manual chaining loop, and its job is exactly the part people screw up when they try to build long videos by hand: the overlap. In a continuation workflow every clip is sampled with context from the previous one, which means each new clip's first frames repeat the tail of the last clip. Finish trims that repeated overlap, sample-aligns the audio, and - critically - captures a CPU-side audio/video latent state you hand to the next Join so the chain can keep going.

It sits after a KSampler in the classic V1 graph: Join (prepares the continuation) → sample → Finish (trims, aligns, captures state) → next Join. The node takes samples (the LATENT you just generated), images and audio (the decoded versions of that segment), and plan (the H3_CONTINUUM_PLAN from the preceding Join). Outputs:

  • images and audio - the trimmed, aligned segment, ready for Assemble to append to the growing result.
  • state (H3_CONTINUUM_STATE) - the continuation state that the next Join consumes as previous_state. This is the actual "continuum" - lose it and the chain is broken.
  • report (STRING) - what got trimmed and aligned, viewable via Core Preview as Text.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum.git

or via ComfyUI Manager ("H3 Continuum"), then restart. No extra pip packages; needs a MiniMax H3 ComfyUI environment (weights not bundled) and ComfyUI v0.32.0+.

Two classic mistakes with this node. The obvious one: dropping the state wire and "finishing" without capturing state - then the next Join starts with no context and you're back to unconnected clips. The subtler one: trimming the latents yourself before Finish expecting a shortcut. The pack's design decodes the full raw chunk first and then Finish trims decoded output, because trimming latents before decode would change the temporal VAE conditioning. If you're starting fresh rather than maintaining a V1 workflow, the V3.3 sampler and Assemble + Seam do all of this internally - Finish is the educational / legacy path.

CategoryMiniMax H3/Continuum/Legacy

Inputs (4)

NameTypeDefaultDescription
samplesLATENT
imagesIMAGE
audioAUDIO
planH3_CONTINUUM_PLAN

Outputs (4)

NameTypeDescription
imagesIMAGE
audioAUDIO
stateH3_CONTINUUM_STATE
reportSTRING