Diffusers Scheduler
The tiny switch that takes MimicMotion from 25 steps down to single digits
- scheduler
Diffusers Scheduler is the pack's optional speed knob. It doesn't generate anything itself - it builds a scheduler object (a DIFFUSERS_SCHEDULER) that you plug into the optional_scheduler input of MimicMotion Sampler to change how that node denoises. Skip it and the Sampler uses its built-in default schedule, which is perfectly fine. Wire it in and you unlock the pack's few-step party trick.
The one choice that matters
scheduler gives you two options:
EulerDiscreteScheduler- the standard SVD schedule. Same family as what the Sampler already uses internally, just with the knobs exposed. If you pick this, you're mostly here for the sigma controls below.AnimateLCM_SVD- this is why the node exists. It's a consistency-model-style few-step scheduler from the AnimateLCM lineage (the LCM-style distillation that got AnimateDiff down to ~6 steps back in 2024). With it, you can drop the Sampler's steps from ~25 to single digits and get roughly comparable output in a fraction of the time. Quality takes a mild hit; speed gains are enormous.
Then the fine print: sigma_min/sigma_max define the noise schedule range (defaults 0.002 and 700 match SVD's training), and align_your_steps swaps in a hardcoded 11-step "Align Your Steps" sigma set tuned for SVD. If you're chasing the fast path, turn it on alongside AnimateLCM_SVD; for one-off default-quality runs, leave everything as is.
What it feeds
The output is a DIFFUSERS_SCHEDULER object - and it only means something inside this pack. It wires into MimicMotion Sampler's optional_scheduler input. It will not plug into a regular KSampler; this is not a general-purpose ComfyUI scheduler node. Don't reach for it outside a MimicMotion workflow.
Install
You get it with the pack - no separate install. ComfyUI Manager → search "ComfyUI-MimicMotionWrapper" → install, or:
cd ComfyUI/custom_nodes
git clone https://github.com/kijai/ComfyUI-MimicMotionWrapper
cd ComfyUI-MimicMotionWrapper
pip install -r requirements.txt
Only three pip dependencies (diffusers, transformers, accelerate) and no models of its own - the scheduler is constructed from config, not weights.
When to bother
Honestly? This is a niche node inside a niche pack. The 2024-era MimicMotion crowd used it to make long dance takes feasible on consumer GPUs, because 25 steps over 200+ frames is an afternoon. If you're just trying the workflow once, leave it disconnected. If you're rendering long clips and patience runs thin, AnimateLCM_SVD plus align_your_steps at ~8 steps is the setting that made this pack usable. That's its entire job, and it does it.
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| scheduler | COMBO | 2 options: EulerDiscreteScheduler, AnimateLCM_SVD | |
| sigma_min | FLOAT | 0.0020–700 | — |
| sigma_max | FLOAT | 700.0000–700 | — |
| align_your_steps | BOOLEAN | false | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| scheduler | DIFFUSERS_SCHEDULER | — |