WanAnimate Sampler Presets

If you've ever stared at a Wan sampler node and thought "is it 20 steps with unipc, or 4 steps with dpm++_sde and CFG 1.0?" - this node is the answer key. WanAnimate Sampler Presets turns the whole steps/cfg/scheduler/LoRA-strength dance into a single dropdown, with a few sane defaults built in and room for your own saved recipes.

It's the other utility node in Magos Digital Studio's ComfyUI-Magos-Nodes pack, alongside Wan Ratio & FPS. Where that one fixes resolutions, this one fixes sampler config. It's a dashboard node: it doesn't run any diffusion itself, it just outputs values you wire into your actual sampler.

How it works

Pick a preset and the node outputs four values: steps (INT), cfg (FLOAT), scheduler (STRING), and lora_strength (FLOAT). The built-ins are:

• Preview / Distilled (4 Steps) - 4 steps, CFG 1.0, dpm++_sde, LoRA strength 1.0. This one is for distilled models (lightx2v-style, CausVid - the few-steps-at-CFG-1.0 crowd). It will produce garbage on the base Wan model.
• Standard (20 Steps) - 20 steps, CFG 5.0, unipc, LoRA 0. The everyday default.
• High Quality (30 Steps) - 30 steps, CFG 5.0, unipc, LoRA 0.
• Maximum (40 Steps) - 40 steps, CFG 5.0, dpm++_sde, LoRA 0.

Pick Custom and you drive custom_steps, custom_cfg, custom_scheduler, and custom_lora_strength yourself. There are 21 scheduler options to choose from (unipc, unipc/beta, dpm++, euler, deis, lcm, the flowmatch family, and so on), so whatever your workflow's model expects, it's probably listed.

The part that's actually nice

Your own presets are saved. The node has a tiny built-in API - a save route at /magos/presets - and user presets persist to user_presets.json across sessions. So once you dial in the combination that works for your LoRA/step-distilled setup, you name it and it shows up in the dropdown forever. Built-ins can't be overwritten, which is sensible. If you want a workflow where you can hand someone a single widget and say "pick the fast one," this is how.

What to wire it into

The outputs go into your sampler - in a Kijai-style WanVideo wrapper, that's the WanVideoSampler node's steps, cfg, and scheduler inputs (and lora_strength into your LoRA loader/apply for the distilled pass). On the base WanAnimate model, Standard is the honest starting point; go Preview/Distilled only when a distillation LoRA is actually loaded. The community's speed-LoRA lesson applies here: distilled settings trade away quality, so don't leave the 4-step preset on for final renders.

Install

Ships in the ComfyUI-Magos-Nodes pack:

cd ComfyUI/custom_nodes
git clone https://github.com/MagosDigitalStudio/ComfyUI-Magos-Nodes

Restart ComfyUI; find it under MAGOS Nodes → Utils. It's a pure-utility node - no Kijai dependency, no models to download.

Common issues

The main trap is conceptual: "Preview / Distilled" is not a "draft mode" for the base model. Four steps at CFG 1.0 on non-distilled Wan gives mushy, uncooked output and then people blame the node. Use Standard for normal runs. The other mild annoyance: your saved presets live in the pack's user_presets.json, so a reinstall wipes them - export your favorites before blowing the folder away.