💊 CR Apply LoRA Stack

Chaining LoRAs the stock way is a wire nightmare - each Load LoRA node threads model and CLIP through to the next, and five LoRAs means five nodes in a daisy chain cluttering your graph. The stack pattern fixes that: you build a list of LoRAs (with their weights) in a separate stack node, and this one applies the whole list to your model and CLIP in a single step.

It's the "apply" half of Comfyroll's LoRA stack system (Suzie1 / RockOfFire). You almost never use it alone - it's the terminal node that takes a LORA_STACK and actually bakes those LoRAs into your model. The famous version of this pattern is rgthree's Power Lora Loader, but Comfyroll's stack nodes do the same job and predate a lot of the copies.

How it works

A LoRA is a small trained adjustment layered onto a base model - it nudges the weights toward a style, character, or concept without you swapping checkpoints. This node takes a LORA_STACK (an ordered list of LoRA + model-weight + clip-weight entries, built by CR LoRA Stack) and applies each one in turn to the incoming model and CLIP, then outputs the modified pair. Order in the stack is the order of application.

The inputs that matter

• model and clip - the pair you're modifying, straight from your checkpoint loader (or the previous stage of your graph).
• lora_stack - the list of LoRAs to apply, produced by a CR LoRA Stack node. This is where the actual LoRA choices and per-LoRA weights live; this node just executes them.

Outputs are the modified MODEL and CLIP, plus the show_help link. Both go downstream exactly where an unmodified model/CLIP would - sampler and text encoders.

How to install it

• ComfyUI Manager: search Comfyroll Studio, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes.git, then restart.

The pack ships no LoRAs - drop your .safetensors files in ComfyUI/models/loras so the stack node can see them.

Common issues

This node needs a lora_stack to do anything - plug it straight into a checkpoint with no stack and it's a no-op passthrough. If your LoRAs aren't showing any effect, check the stack node upstream, not this one.

The usual LoRA headaches apply. Weights stack: three LoRAs at 1.0 each will fry your image into an oversaturated mess, so pull them down (0.5–0.8 each is a saner starting range when you're combining several). Mismatched base architecture is the other classic - an SDXL LoRA on an SD 1.5 model, or vice versa, either errors out or silently does nothing useful; the LoRA has to match your checkpoint's family. And if you stacked LoRAs that fight each other (two strong style LoRAs, say), lowering weights beats removing one - blend, don't brawl. If the node itself is missing, that's the Comfyroll pack failing to load, a pack-level issue rather than this node.