Register LoRA Hook (Model Only) 🎭🅐🅓

This is the model-only cut of ADE_RegisterLoraHook - same LoRA Hooks registration, minus the CLIP input and its strength control. If you've used ComfyUI's core LoraLoaderModelOnly relative to LoraLoader, it's the exact same relationship, just applied to this pack's hookable LoRA system instead of a plain loader.

What it does

A LoRA Hook is a LoRA that stays inert until you explicitly attach it - to a piece of conditioning, a masked region, or a scheduled time window - rather than applying globally the instant it's wired in, which is how a normal loader behaves. This node registers a LoRA as one of those hooks, applying it to model only and skipping the text encoder entirely. Reach for it when the LoRA you're hooking is motion- or structure-focused with a negligible CLIP-side component, or when you're handling text conditioning through a separate path and don't need this node touching it at all.

Inputs and output

Required: model (MODEL), lora_name (a dropdown of your LoRA files), strength_model (default 1, range −20 to 20).

Two outputs: MODEL - the hooked version (route your unhooked base model to anything that shouldn't carry this effect, the same caution that applies to the full ADE_RegisterLoraHook) - and HOOKS, the actual payoff: a handle you feed into ADE_AttachLoraHookToCLIP or a conditioning-side attach node to gate where this LoRA's effect actually fires.

How to install it

ComfyUI Manager: search AnimateDiff Evolved, confirm the author is Kosinkadink, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved

You need the LoRA file itself in your normal LoRA folder, same as any standard loader. No AnimateDiff-specific download tied to this node.

Common issues & troubleshooting

Expected CLIP-side changes and don't see them. That's correct - this variant never touches the text encoder. If your LoRA has a meaningful CLIP component you're relying on, use the full ADE_RegisterLoraHook instead, which applies to both.

LoRA seems to apply everywhere, not just where you attached the hook. Check that you're routing the unhooked MODEL output - or your original checkpoint's model - to any part of your graph that shouldn't carry this effect. The hook mechanism only gates things downstream of an actual attach node; wiring this node's hooked MODEL straight into your sampler applies it unconditionally regardless of any hook.

Not sure what to do with HOOKS. It does nothing on its own. You need a follow-up node like ADE_AttachLoraHookToCLIP or ADE_AttachLoraHookToConditioning to actually tie it to a piece of conditioning before it has any effect - this node is only the registration step.

Wondering if this is overkill. For most AnimateDiff work, yes. If you just want a LoRA active for the whole clip, LoraLoaderModelOnly is simpler and does the job. Reach for hooks specifically when the LoRA needs to apply only to part of a scheduled prompt sequence or only during certain frames.