Set CLIP LoRA Hook 🎭🅐🅓

Displayed as Set CLIP LoRA Hook, this is the second half of the pack's LoRA Hooks system, and it's the piece that actually gates a LoRA's effect to something specific. ADE_RegisterLoraHook builds a HOOKS handle representing "this LoRA, at this strength" - inert on its own. This node attaches that handle to a CLIP model, so any text you encode through the resulting hook_CLIP carries the LoRA's influence, while text encoded through your original, unhooked CLIP doesn't.

That's the whole mechanism for getting a LoRA to apply to some of your conditioning and not all of it - say, a style LoRA you want on your main subject prompt but not on a background or negative prompt sharing the same graph.

How it works

CLIP text encoding is where prompt text becomes the conditioning a sampler actually uses. By attaching a lora_hook to the CLIP model before encoding, this node makes that specific encode pathway carry the hooked LoRA's effect, while any other CLIPTextEncode in your graph using the original, unhooked CLIP output stays unaffected. It's a narrower, more targeted alternative to ADE_RegisterLoraHook's own CLIP output, which is hooked but not tied to anything specific yet.

The inputs and outputs that matter

Both required, nothing optional: clip (the CLIP model you want to attach the hook to - typically your base, unhooked CLIP from your checkpoint loader) and lora_hook (type HOOKS, built upstream by ADE_RegisterLoraHook). Output is a single hook_CLIP, which you feed into a CLIPTextEncode exactly like any normal CLIP model - the hook rides along invisibly and only affects that specific encoding.

How to install it

Standard for the pack - ComfyUI Manager, search AnimateDiff Evolved by Kosinkadink, or:

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

No model download tied to this node itself. You need ADE_RegisterLoraHook upstream, and whatever LoRA file that node is registering, in your normal LoRA folder.

Common issues & troubleshooting

Not sure what to plug into lora_hook. This node is useless without ADE_RegisterLoraHook feeding it first - that's where the HOOKS object actually gets built from a LoRA file and strength values. If you landed on this page looking for a standalone LoRA node, this isn't it; look at ADE_RegisterLoraHook for the piece that loads the LoRA.

LoRA effect shows up everywhere, not just on the prompt you attached it to. Double-check you're routing the hook_CLIP output specifically into the CLIPTextEncode for the prompt you want affected, and your original unhooked CLIP into everything else. If every CLIPTextEncode in the graph is pulling from hook_CLIP, the LoRA is effectively global again - the hook only narrows scope if the wiring actually reflects that intent.

Effect isn't showing up at all. Make sure the CONDITIONING produced from this hooked encode is actually the one reaching your KSampler - if something downstream overwrites or replaces it (a scheduling node, a conditioning-combine step), the hook goes with it silently.

This is a narrow, deliberately advanced feature. Most AnimateDiff workflows never need per-prompt LoRA gating - reach for it specifically when a single global LoRA application genuinely isn't what your prompt structure calls for.