Easy Apply LoraStack

Running four LoRAs at once usually means four LoRA Loader nodes chained model → model → model → model, and a canvas that looks like a switchboard. easy loraStackApply cleans that up. You build a stack once (with easy loraStack), then this node applies the entire stack to your model in a single step.

The split - one node to define the stack, another to apply it - is what makes it nice. You can pass that stack around, feed it into a loader, swap the whole set out at once, or fork it. It's the same pattern the pack uses for controlnet stacks.

How it works

A LORA_STACK is just a list of (LoRA name, model strength, clip strength) entries. loraStackApply walks that list and patches each LoRA onto your model in order, exactly as if you'd chained individual loaders - same result, one node. If you also give it a CLIP, it patches the text-encoder side of each LoRA too, which matters for LoRAs that carry trigger-word conditioning.

The inputs and outputs that matter

• lora_stack (LORA_STACK) - the stack to apply. Build it with easy loraStack (or any node that emits a compatible stack) and wire it in here.
• model (MODEL) - the model to patch. Comes out modified.
• optional_clip (CLIP) - optional but recommended. Without a clip, only the model (UNet) side of each LoRA is applied; with it, the clip side is patched too. Many LoRAs behave noticeably better when the clip side is included.
• Outputs: model and clip - feed both onward to your conditioning + sampler.

How to install it

ComfyUI Manager: search ComfyUI-Easy-Use, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

then install.bat / pip install -r requirements.txt, restart. No extra models - it uses the LoRA files already in your models/loras folder, referenced by the stack.

Common issues & troubleshooting

Trigger words aren't firing. If a LoRA needs a keyword to activate its concept and it's doing nothing, check that you passed optional_clip. Skipping the clip means the text-encoder half of the LoRA never gets applied, and trigger-word LoRAs lean on exactly that.

Too many LoRAs, muddy output. Stacking is convenient but it doesn't repeal the usual rule - pile on too many strong LoRAs and they fight, wash each other out, or blow up the image. Pull the per-LoRA strengths down in the stack; total effective strength across the stack adds up faster than you'd think.

Order can matter. LoRAs apply in stack order, and for some combinations the order shifts the result. If a stack looks off, try reordering the entries in easy loraStack.

"clip is None" downstream. If you didn't wire optional_clip, the node's clip output has nothing to pass. Either provide a clip, or take the clip for your CLIP Text Encode from your loader directly instead of from this node.