MiniMax H3 Easy Segment Sample

MiniMax H3 Easy Segment Sample (class MiniMaxH3EasySegmentRender) is the node that turns your segment plan into actual video. Context Segments up the chain builds the multi-shot plan with prompts, durations, and continuity settings; this node walks that plan shot by shot, runs the sampler on each one, and hands you the first-pass result. If Context Segments is the director's script, this is the camera crew.

What it needs

Its inputs are refreshingly minimal - this is where the pack hands control back to ordinary ComfyUI:

• h3_context (MINIMAX_H3_CONTEXT) - the plan from MiniMax H3 Easy Context Segments.
• model (MODEL) - the H3 model, straight from the Context Segments node's model output (run it through LoRAs or patches first if you want; the README's whole point is that samplers and patches stay ordinary connections).
• sampler (SAMPLER) and sigmas (SIGMAS) - the sampler settings, exactly as you'd pick them on a KSampler. If you're not sure, pull the same sampler/scheduler combo the official H3 nodes use.
• seed - your seed, with the usual control_after_generate widget.

The single output, segments (MINIMAX_H3_SEGMENTS), feeds Segment Decode for a first-pass preview or Segment Refine for the resolution-aware second pass.

How it works

The node iterates the plan's shots in order. For each one it rebuilds that segment's own conditioning (its prompt and its @-referenced media - nothing from the shared library that the segment didn't explicitly reference), then samples with continuity from the previous segment's tail. The continuity mechanism follows whatever continuity_mode the Context Segments node chose: a latent pass-through, a re-encoded RGB guide, or an AV prefix with its nested video/audio denoise mask so the sampler preserves the copied prefix instead of regenerating it.

It reports progress per shot through ComfyUI's progress bar, so on a long plan you can see exactly which segment it's chewing on. Nothing about this stage changes resolution - it samples at the canvas set in the Context Segments node. Refinement is a separate, later stage on purpose.

The gotcha that will annoy you

Segment Sample re-runs on every queue, even with an unchanged seed and inputs - the node deliberately reports itself as always-changed. That's by design (the author wants every queue to be a fresh segment run), but it means no node-cache fast path: if you tweak one thing and re-queue, every segment resamples, not just the changed one. So do your prompting and continuity tuning on a short two- or three-segment plan before committing to a ten-shot render. The flip side is that this makes iteration predictable - no stale cached segments silently tainting the chain.

Installing

This ships in ComfyUI-MiniMaxH3-Easy (same install as the rest of the pack). Update ComfyUI to a build with the official MiniMax H3 nodes first, then install via ComfyUI Manager (search ComfyUI-MiniMaxH3-Easy, pick the Nightly version - the README warns the published tags can lag the repo) or:

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

Restart ComfyUI. The easiest way to see the whole thing wired correctly is to import workflow/MiniMax_H3_Easy_Context_Segments.json from the repo and read the connections - Context Segments → Segment Sample → Segment Decode is the core loop you'll be living in.