H3 Continuous - Output / Stitch (Legacy v1.0)
H3 Output / Stitch v1.0 — Full, or Stitch Ready? Learn This or Suffer.
- images
- handover
- audio
- images
- audio
- trim_info
- head_trim_frames
- tail_trim_frames
Every H3 clip in a continuation chain needs to come out in one of two flavors, and getting them mixed up is the fastest way to ruin a stitch. H3ContinuousStitchOutputV1 is the node that decides, and the decision is the whole job: Full or Stitch Ready.
- Full is a true bypass - you get the complete render, exactly as sampled. Use it when you want to inspect a clip or when it's genuinely the final output with its full landing intact.
- Stitch Ready is the secret sauce. It removes the reused head context (that's the duplicated history a Continue node carried into clip 2+, defaulting to
head_context_frames) and trims the tail after this clip's exact phase-aligned latent handover boundary - the frozen FL2VA landing the Auto Handover analyzer marked as dead. What's left is a clip whose end lines up frame-exactly with where the next clip's carried context begins. That's what makes direct concatenation possible without double motion or a frozen tail bleeding into the next segment.
So the workflow rule is simple: every intermediate clip is Stitch Ready; the final clip of the project is Full (or Stitch Ready if you want the safety trim applied to the very end - the author documents that choice explicitly in the newer versions).
Inputs and outputs
Required: images (the decoded full render) and output_mode. Optional: handover - connect the Auto Handover analyzer's output; required for Stitch Ready so the tail trim matches the exact latent cutoff; audio (the decoded audio, passed through and trimmed in step); head_context_frames - leave unconnected (0) for Clip 1, connect actual_head_context_frames from the Continue node for Clip 2+.
Outputs: images, audio, trim_info (a string explaining what was cut), and the two ints that tell you exactly what happened: head_trim_frames and tail_trim_frames. Those two numbers are your sanity check - if Stitch Ready trimmed zero tail frames, your handover input probably wasn't connected.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/HerrgottMargott/Herrgotts-H3-Infinite-Continuation-Suite.git
Restart ComfyUI, reload. No pip deps, no bundled weights - H3 models/VAEs from Comfy-Org's MiniMax-H3 repo.
Gotchas
This is the v1.0 class, and it's the parent of the v1.2/v1.4 Stitch nodes - the newer ones add a Final Clip mode and v1.4's shared-boundary geometry. If you're on an old v1.0 chain, remember the handover is required for Stitch Ready to mean anything; without it, the node has no idea where the dead tail starts and will trim by guesswork. And never feed the Stitch Ready output back into the next Continue's previous_latent - continuation needs the full latent, not the trimmed render.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | Decoded FULL rendered frames for the current H3 clip. | |
| output_mode | COMBO | Full | Full is a true bypass. Stitch Ready removes the reused head context (clip 2+) and the tail after this clip's exact phase-aligned latent handover boundary. |
| handoveropt | H3_CONTINUOUS_HANDOVER | Connect the current clip's Auto Handover Analyzer output. Required for Stitch Ready so the tail matches the exact latent cutoff. | |
| audioopt | AUDIO | — | |
| head_context_framesopt | INT | Clip 1: leave unconnected (0). Clip 2+: connect actual_head_context_frames from H3 Continuous - Continue from Latent v1.0. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| audio | AUDIO | — |
| trim_info | STRING | — |
| head_trim_frames | INT | — |
| tail_trim_frames | INT | — |