H3 Studio · Director Sampling Preset

If you've used the H3 Studio · Image Director, you know its Speed/sampling dropdown - that 11-choice menu from base quality through LightX and PDD. H3StudioContextSamplingPreset is how you get that same decision, made in one place, applied inside a workflow or subgraph where you're not using the Director's built-in path. It's the "Director-aware" cousin of the plain H3 Studio · Sampling Preset: same shape of outputs, but it reads the sampling intent from the studio_context instead of making you re-pick everything.

The inputs are just two connections: the loaded H3 diffusion model and the studio_context from the Director. That's it - no profile dropdown to disagree with what you already chose. The node resolves which recipe the Director's context is asking for and returns model, sampler, sigmas and sampling_info, ready for a custom sampler.

The delegation detail

The interesting mechanism here is how it handles the different profile families. The description spells it out: base and LightX recipes stay fully local - the node builds those sampler/scheduler/sigma combinations itself. But Mamad8 PDD profiles delegate to the separately installed external node package (ComfyUI-MiniMaxH3-PDD-Mamad8) and match it with the local PDD student LoRA and heads. If you pick a PDD profile and the external package isn't there, the node fails with corrective instructions rather than silently doing something wrong - and, per the pack's own docs, it "never alters Base behavior." A missing PDD dependency shouldn't break your default profile.

That PDD path is worth understanding because it's the one part of this pack that isn't self-contained. PDD is a four-step REF2VA-only accelerated path, and it needs both the separate GPL node package and the matching LoRA + heads files. If you're not doing reference edits, you don't need any of it.

Why use this instead of the plain preset

In a hand-built graph you could absolutely just wire the standalone Sampling Preset and pick a profile yourself. But if you're building a workflow that's driven by the Director (multiple routes, configurable references, possibly a shared subgraph), this node means the sampling matches whatever the Director decided - one source of truth instead of two dropdowns that can drift apart. The sampling_info output tells you which recipe actually got applied, so you can verify the delegation happened the way you expected.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/thaakeno/ComfyUI-MiniMax-H3-Studio.git
cd ComfyUI-MiniMax-H3-Studio
python -m pip install -r requirements.txt

Restart and hard-refresh the frontend. Only add the external PDD package if you actually plan to use PDD profiles. Otherwise the core pack has no extra dependencies.