MiniMax H3 Easy Segment Collect

MiniMax H3 Easy Segment Collect is the capstone of the pack's per-segment workflow, and it's genuinely easy to misunderstand, because you never connect it to most of the chain. You connect it to one node - the last Segment Step - and it does all the bookkeeping.

The context: H3 is a 33B open-weight video model, and MiniMax's own ecosystem answer to "I want more than a single 5-second shot" is chaining shots that hand context to each other. This pack's normal one-node path, MiniMax H3 Easy Segment Sample, generates every segment in one go - but it re-executes on every queue even when nothing changed. When you want independent per-segment seeds, per-segment prompt overrides, and selective reruns that ComfyUI can actually cache, you replace that single node with a chain of Segment Step nodes (the pack's 7.MiniMax_H3_Easy_Context_Segments_Control.json example). Each Step samples one shot and passes its accumulated result to the next through Previous segment.

Segment Collect is what sits at the end of that chain and converts it back into the same MINIMAX_H3_SEGMENTS result that Segment Sample would have produced - the type Segment Decode and Segment Refine expect. Think of it as a reducer: it walks the chain you built and folds it into one finished, decodable segment list.

How it works

Feed it only the final Step. The node checks that this Step's index equals the total number of shots in the plan - if you connect segment 2 of 3, it errors and tells you exactly that it needs the final one. It then walks backward through every Step's internal previous pointer, guarding against cycles, verifies the chain is continuous and belongs to one plan, and reverses it into order. Each Step's sampled prompt and referenced media get merged back into its shot entry, and the whole plan is returned as a single segments result.

That check-first design is the whole point. A mis-wired chain fails loudly here instead of silently producing a jumbled video.

The inputs and outputs that matter

This one is almost too simple to need a list, but here it is:

• final_segment (required) - the output of the last MiniMax H3 Easy Segment Step.
• segments (output) - a MINIMAX_H3_SEGMENTS result, wired into MiniMax H3 Easy Segment Decode for a preview, or Segment Refine if you want a per-segment second pass first.

The wiring pattern is:

Context Segments → Sample Setup → Segment Step 1 → Segment Step 2 → … → Segment Step N
                                                                              ↓
                                                                  Segment Collect → Segment Decode

Installing it

It ships in the ComfyUI-MiniMaxH3-Easy pack:

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

Then restart ComfyUI. Via ComfyUI Manager, search ComfyUI-MiniMaxH3-Easy and choose the Nightly release - the others can lag the repo. You'll need a current ComfyUI with the official MiniMax H3 support and the H3 model files in the usual ComfyUI/models folders before any of this runs.

Common issues

The two failure modes are wiring the wrong Step and forgetting the chain boundary. Collect refuses anything that isn't the final segment, and a chain with more Steps than the plan has shots fails at the Step level before you even get here. One scope note worth knowing: the per-step control chain exists for text-driven Context Segments generation. If your plan came from MiniMax H3 Easy Selected Video Context, that path goes through Segment Sample instead - Sample Setup rejects selected-video contexts, so don't try to force one through a Step chain.