Diffusers Scheduler Loader &MC

Diffusers Scheduler Loader &MC is where you pick the sampler for the pack's "&MC" diffusers pipeline line. It's a small node with one job - load a scheduler and hand it to the next stage - but it's also the one that reminds you this side of the pack is a diffusers pipeline wearing ComfyUI clothes, because it speaks fluent HuggingFace scheduler names instead of ComfyUI's euler / uni_pc vocabulary.

It sits in the middle of the diffusers chain:

Load Magic Clothing Pipeline → Change Pipeline Dtype And Device →
Diffusers Scheduler Loader &MC → Diffusers Model Makeup &MC →
Load Magic Clothing Adapter → RUN Magic Clothing Diffusers Model

What it takes and gives

• pipeline (PIPELINE) - the loaded pipeline object. It needs this because the scheduler is constructed from the pipeline's own config - the node reads the scheduler config that came with the model and swaps the algorithm on top of it.
• scheduler_name - one of eleven diffusers schedulers: DDIM, DDPM, DEISMultistep, DPMSolverMultistep, DPMSolverSinglestep, EulerAncestralDiscrete, EulerDiscrete, HeunDiscrete, KDPM2AncestralDiscrete, KDPM2Discrete, or UniPCMultistep.

Output is a single SCHEDULER object, which wires into Diffusers Model Makeup &MC - that's the node that actually installs it into the pipeline (pipeline.scheduler = scheduler) and swaps in your chosen VAE. The loader alone doesn't change anything about the pipeline; if you skip the makeup step, your new scheduler never takes effect.

Which one should you pick

For virtual try-on at a sane step count, UniPCMultistep is the pack's own default (the OMS-era pipeline used it explicitly) and is a safe bet at 20–25 steps. EulerDiscrete is the fast-and-fine choice, DPMSolverMultistep is the quality pick if you can afford the steps, and DDIM is there if you want the classic stable-diffusion feel. The ancestral variants (EulerAncestralDiscrete, KDPM2AncestralDiscrete) add noise-based variety per step, which can help break the flat look but makes results less reproducible at a fixed seed.

Install

Same as the rest of the pack - Manager, search "comfyui-magic-clothing", or:

cd ComfyUI/custom_nodes
git clone https://github.com/longgui0318/comfyui-oms-diffusion
# restart

The one real gotcha here is that this node - like every node in the diffusers line - imports diffusers at module load, so the package has to be in your ComfyUI Python environment. And the scheduler name list is fixed; if your scheduler of choice isn't in the eleven, you can't type it in. The pragmatic advice: don't overthink this one. Magic Clothing's success rate is the bottleneck, not the sampler, so pick UniPCMultistep or Euler and spend your tuning effort on cloth_guidance_scale instead.