XY Input: LoRA Stacks

Most of the LoRA XY plotters sweep one LoRA's model or clip strength. This one is coarser and more interesting: it plots whole LoRA stacks against each other. Feed it two to five complete stacks - each potentially several LoRAs deep - and it renders one column or row per stack, so you can compare entire combinations side by side instead of one weight at a time. It's part of the Efficiency Nodes XY Plot system.

Where you'd reach for it: you've built "anime style + detail enhancer" as one stack and "anime style + different detail LoRA" as another, and you want to see which combination wins on the same prompt and seed. Comparing multi-LoRA recipes by eyeballing separate runs is miserable; this settles it.

How it works

Each lora_stack_N input takes a LORA_STACK - the bundle produced by a LoRA Stacker node (the pack has its own, and other packs like rgthree's Power Lora Loader emit the same type). The node lines those stacks up as the values along one plot axis and outputs a single X or Y connection into the XY Plot node.

node_state is the only widget and it's essentially just an enable flag. The real content comes from the stacks you wire in.

The inputs and outputs

• node_state (required) - leave it Enabled.
• lora_stack_1 … lora_stack_5 (optional) - the stacks to compare. Connect at least two for a meaningful plot; unused slots are ignored.

Output is X or Y (type XY), which goes to the XY Plot node's X or Y input. That node plugs into an Efficient KSampler's script input.

Installing it

ComfyUI Manager → search Efficiency Nodes for ComfyUI, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/jags111/efficiency-nodes-comfyui

Restart ComfyUI. Your LoRA files live in models/loras as usual; you build the stacks with stacker nodes and route them here.

Common issues

The one that's specific and worth knowing: the community established years ago that in the Efficiency SDXL base+refiner flow, a LoRA stack applies to the base only, not the refiner. Someone tore the setup apart and confirmed the LoRA never reaches the refiner stage. So if you're plotting LoRA stacks on an SDXL workflow that also runs a refiner, be aware the refiner pass is seeing an un-LoRA'd model - your comparison is really a base-model comparison. On a single-stage (base-only) workflow this doesn't bite.

The standard XY Input caveat also holds: the output does nothing until it flows through the XY Plot node into one of the pack's Efficient KSamplers. A stock KSampler won't run it.

If a stack seems to have no effect, check that the stacker feeding it actually has LoRAs selected with non-zero strengths - an empty stack plots as a blank variation rather than erroring.

Pack-wide: an IMPORT FAILED at startup (commonly a pip freeze non-zero exit) takes the whole pack offline. Update ComfyUI, update the node, check the environment.