Lora Loader Model Only XY 🍌

"What strength should I run this LoRA at" is one of those questions you answer by just looking at a grid of outputs, not by guessing. This node is the setup half of a mini XY-plot system built into the pack: instead of loading a LoRA at one fixed strength, you give it a list of strengths and it prepares everything needed to sample all of them in one batch. Pair it with KSampler XY and you get one output per strength, side by side, without manually cloning your loader node ten times.

How it works

You pick one LoRA and type out the strengths you want to test as a plain list of numbers. The node doesn't apply the LoRA itself - it packages the model, the LoRA choice, and the strength list into two special types (XY_MODEL and XY_LIST) that only this pack's KSampler XY node understands. KSampler XY then loops the actual sampling across each strength for you.

Inputs and outputs that matter

• model - the base model to test the LoRA against.
• lora_name - pick from your installed LoRAs, same as any normal loader.
• strength_list - a multiline text field where you type the strengths to sweep, e.g. one per line or comma-separated (match whatever the field's own formatting expects). This is the whole point of the node: instead of one strength value, you get a batch of them.
• Outputs: XY_MODEL and XY_LIST, both of which only make sense wired into KSamplerXY|cgem156 - they won't connect to a regular KSampler or a regular LoRA loader downstream.

Note the "model only" in the name: this loader patches the model but does not touch CLIP/text conditioning, same distinction as the stock LoraLoaderModelOnly node.

Installing it

ComfyUI Manager → search cgem156-ComfyUI🍌 → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/laksjdjf/cgem156-ComfyUI

then restart ComfyUI. No extra downloads required for this node.

Common issues & troubleshooting

Nothing to connect it to. If you try to wire XY_MODEL or XY_LIST into a normal KSampler, it'll refuse - these are custom types built specifically for this pack's own KSampler XY node, not drop-in replacements for the standard sampling pipeline. Bring both nodes in together or this one is dead weight.

Strength list not being parsed the way you expect. Since it's a free-text multiline field rather than a structured list of number widgets, formatting matters. If your sweep comes out shorter or longer than you typed, double-check the separator you used matches what the field expects (a blank line or stray character is an easy way to get an extra "strength" you didn't intend).

Wasted compute on a wide sweep. Ten strengths means ten full samples. If you're just narrowing in on a rough range, start coarse (0, 0.5, 1) before spending a full run on a fine-grained sweep across every 0.1 step - same logic as any parameter search, just applied to LoRA strength instead of CFG or steps.

Confusing this with the real weight-setting node. This is a testing tool. Once you've picked your strength from the grid, go back to a normal LoRA loader for production runs - there's no reason to carry the XY plumbing into a workflow that only ever runs one strength.