H3 Infinite - Output / Stitch — Masked AV v1.4

The v1.4 version of H3 Infinite - Output / Stitch does exactly what the v1.2 node does - takes a rendered H3 clip and cuts it into Full, Stitch Ready, or Final Clip form - but it's tuned for the suite's newer native Masked AV continuation path. If you're on a current ComfyUI build with MiniMax H3 mask support, this is the node you want, because it understands where the v1.4 handover boundary actually is. And in v1.4, that boundary means something slightly different than it used to.

What changed in v1.4

Older guide-based continuation reconstructed the previous clip through keyframe rows and could let other conditions nudge the opening frames. The v1.4 path instead copies the previous clip's full AV latent directly into the new target and protects it with ComfyUI's native per-stream denoise masks. One consequence: the visible video and the next clip's protected context now share one exact boundary. Auto Handover picks a freeze/brightness-safe visual endpoint, then snaps it backward to the latest boundary that's exact for both the video and audio latents. This node's Stitch Ready mode trims each clip to that same shared boundary, so what you see end and what the next clip reuses are the same frames. No temporal mismatch between the two - that mismatch was a big source of visible jumps in older approaches.

The other v1.4 twist is that audio is allowed to outlive the picture. Because the video and audio masks are independent, the next clip can keep protecting the previous clip's audio tail even after the video handover cut. This node removes the normal reused video head (via actual_head_context_frames), and the carried-over protected audio tail simply survives on the next clip after that trim. Dialogue that runs past the visual cut isn't forced to die at the seam anymore.

Inputs and outputs

Identical layout to v1.2:

• images - the decoded FULL render for the current clip.
• output_mode - Full (true bypass), Stitch Ready (removes reused head + freeze-safe tail, for intermediate clips), or Final Clip (removes only the reused head, keeps the complete Last Frame landing for the last segment).
• handover - required for Stitch Ready tail trimming; the analyzer metadata that says where the shared boundary is. Final Clip ignores the tail cutoff, but you can still connect it so the metadata gets saved for later.
• head_context_frames - leave unconnected for Clip 1; for Clip 2+ connect actual_head_context_frames from the v1.4 Continue node. With the default 39-frame context, that's the number of protected duplicate frames to strip.
• audio - optional, trimmed to match.

Outputs are the rendered images, matching audio, plus trim_info, head_trim_frames and tail_trim_frames if you want to see (or reuse) exactly how much it cut.

The catch: you need a current ComfyUI

The v1.4 nodes require the native MiniMax H3 AV-mask support from ComfyUI PR #15375. The suite deliberately doesn't ship a compatibility shim - it probes the live H3 mask capabilities at runtime and, if masked AV support is missing, it tells you to update rather than half-working. If you get that message: update ComfyUI, fully restart, reload the browser UI. The old v1.2/v1.3 nodes stay registered so your legacy workflows keep loading, but the shipped v1.4 workflows use this node.

Install and gotchas

Install the whole pack via ComfyUI Manager (search Herrgotts-H3-Infinite-Continuation-Suite) or:

cd ComfyUI/custom_nodes
git clone https://github.com/HerrgottMargott/Herrgotts-H3-Infinite-Continuation-Suite.git

Restart after. No extra pip dependencies beyond ComfyUI's own stack, and the MiniMax H3 model weights come from Comfy-Org's repo (the pack has none). One thing to remember if you're new to the suite: the 39-frame protected head still exists internally even though Net New Content compensates for it in the Continue node - which is precisely why this node's head trim matters. Skip it and every seam doubles up 39 frames of your video.