Wan Motion Scale Advanced (Experimental)

The basic WanMotionScale from the comfyUI-LongLook pack gives you a motion-speed dial. This one gives you the same dial plus a theta knob, and the README is refreshingly honest about it: "work in progress - not yet fully documented." If you just want to speed up or slow down motion, skip this and use the basic node - it's simpler and does the job. Reach for WanMotionScaleAdvanced only if your actual goal is getting a single >81-frame generation to stay coherent, because that's what the extra knob is aimed at.

The mechanism builds on the same trick as the basic node. It scales temporal RoPE positions via ComfyUI's rope_options (scale_t, scale_y, scale_x behave identically to the basic node), so the speed control part is a solved problem. What's new is theta, the RoPE base frequency. Wan 2.2 was trained with theta = 10000, and this node lets you mutate the model's rope_embedder.theta directly at load time.

Why would you? Higher theta (think 20000–50000) effectively stretches how far the position encoding can reach, so frames read as "closer together" - the author's hope is that this buys a longer effective context and keeps 81+ frame single generations from falling apart. Lower theta (2000–5000) pushes frames apart temporally, which can amp up motion intensity but also invites instability. Both directions are speculation dressed up as a tooltip, which is fine - it's labeled experimental, the tooltip literally says "start with small changes."

The inputs are model (in), enabled (on by default), scale_t, scale_y, scale_x, and theta (default 10000, and leaving it there means no theta mutation happens at all). The single output is the patched MODEL, wired straight into your sampler like any patcher node. There's a Wan Motion Scale Demo.json example in the pack's examples folder if you want a starting graph.

One honest warning before you go wild: mutating theta changes fundamental model behavior, not a parameter you're supposed to tune per-scene. The enabled toggle is your friend - keep it on the graph for A/B tests, because you will want to see the before and after. And since this node shares the pack with the rest of the LongLook family, installation is identical:

cd ComfyUI/custom_nodes
git clone https://github.com/shootthesound/comfyUI-LongLook.git

then restart ComfyUI, or install via ComfyUI Manager by searching "comfyUI-LongLook." No requirements.txt, no model downloads - it patches whatever Wan 2.2 checkpoint you load, works with i2v and t2v. If theta does nothing on your model, check the console: the node logs whether it found the rope_embedder to mutate. Not every Wan variant exposes it, and the code is written to fail soft rather than crash.

In short: it's a tweaker's node. If you're chasing motion speed, use the basic one. If you're one of the people trying to stretch a single Wan generation past 81 frames without chunking, this is a legitimately interesting thing to experiment with - just don't expect magic.