MiniMax H3 Easy Sample Setup

MiniMax H3 Easy Segment Sample Setup exists purely to keep your graph tidy, and it does that by being the one node in the per-segment workflow that actually has the full sampler plumbing on it. Every sampling call in ComfyUI needs four things: the model, the context to condition on, a sampler, and a noise schedule (SIGMAS). In the one-node MiniMax H3 Easy Segment Sample path, those all live on that single fat node. The moment you break generation out into a chain of MiniMax H3 Easy Segment Step nodes so each shot gets its own seed, you'd normally be forced to repeat all four connections on every Step. Sample Setup collects them once and hands the bundle to the first Step.

That's the whole job. It's a pass-through - an inert package, not a sampler.

How it works

You connect h3_context, model, sampler, and sigmas to it, and it returns a single setup handle. The node validates as it packs: it confirms the context actually carries a segment plan, and it rejects plans built by MiniMax H3 Easy Selected Video Context - those go through Segment Sample, not through the per-step chain, so this node is explicitly for text-driven Context Segments generation. The actual sampling never happens here. At queue time, the first Segment Step consumes the setup, resolves which shot it is, and does the real Guider_Basic sampling with the model, sampler, and sigmas you stored. Later Steps inherit the same setup through the chain without you re-wiring anything.

The inputs and outputs that matter

• h3_context - the H3 Context output of MiniMax H3 Easy Context Segments (not the main Easy node's context). The error message when you get this wrong is unusually helpful: "Connect the H3 Context output from MiniMax H3 Easy Context Segments."
• model - a MODEL, normally from MiniMax H3 Easy Loader (optionally through LoRAs or a Model Adapter).
• sampler - a SAMPLER, e.g. from KSamplerSelect. The example workflow uses euler.
• sigmas - the schedule, e.g. from BasicScheduler (the control example runs 4 steps, beta).
• Output: setup, which connects only to the first Segment Step.

Wiring order in the pack's control workflow:

Context Segments ─h3_context─┐
Model (Loader → LoRA) ──────┤
KSamplerSelect ─────────────┼─→ Sample Setup ─setup→ Segment Step 1 → Step 2 → …
BasicScheduler ─────────────┘

Installing it

Part of the ComfyUI-MiniMaxH3-Easy pack:

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

Restart after installing. In ComfyUI Manager, search the pack and install the Nightly build, since the other published versions trail the repo. You'll also need a reasonably recent ComfyUI that includes the official MiniMax H3 nodes (comfy_extras.nodes_minimax_h3), plus the H3 weights the Loader picks from ComfyUI/models/diffusion_models, text_encoders, and vae.

Common issues

The recurring beginner mistake is treating the setup node as a terminal. It isn't - if you queue with nothing after it, nothing samples, because the setup is only consumed by a first Segment Step. The other classic is connecting setup to the wrong Step: it must feed the first one, and the node raises if you attach it anywhere later in the chain. And if you're coming from the Selected Video Refine workflow, stop: that path bypasses Sample Setup entirely, and the validation error you'll get ("supports generated Context Segments, not selected-video preparation") is telling you to use Segment Sample instead.