Lora List

Comparing three LoRAs means swapping a file in the LoraLoader and re-running three times, hoping you remember which render came from which. Lora List turns that into a dropdown: pick up to six LoRAs once, feed it the grid's current index, and it hands back a model and CLIP with the right LoRA applied - automatically, per cell. It's the pack's most interesting node because it doesn't just select data; it actually performs a model operation.

How it works

Under the hood it's a LoraLoader with extra steps. You give it a base model and clip, and up to six LoRA filenames (lora1 through lora6, dropdowns populated from your models/loras folder with a "None" option). The index picks which one is active. The node then loads that LoRA's weights and applies them with strength_model and strength_clip, exactly like the built-in loader, and returns the patched current model and current clip out the other side.

There's a small cache so the same LoRA isn't re-read from disk every execution - it hangs on to the last one loaded until index moves somewhere else. And two shortcuts worth knowing: selecting "None" passes your model through untouched, and so does setting both strengths to 0.0. That makes the node a clean pass-through when the grid row calls for "no LoRA," which is exactly the baseline you want in a comparison grid.

The inputs and outputs that matter

• index - which slot to use (INT, min 1, 1-based)
• model, clip - the base model and its CLIP, straight from your CheckpointLoader or UNETLoader
• lora1 … lora6 - dropdowns of everything in your loras folder
• strength_model, strength_clip - both FLOAT, default 1.0, range −20 to 20. These are the same strengths the standard LoraLoader takes, so treat them the same way: 0.6–0.8 is a common starting range for style LoRAs, and higher values on unstable models are how you get artifacts.

Three outputs: current model, current clip, and lora name list - that last one is a STRING_LIST of all six names, handy if you want the active LoRA's name to show up in image metadata or a grid label.

How you'd actually use it

Wire it into a Create Image Grid loop: y_index into index, and the loop does six LoRA variants for you while a Save Image Grid collects them into one labeled sheet. For a true side-by-side you'll want the same seed across the row - lock the KSampler seed so the only thing changing between cells is the LoRA. That's the "change one variable at a time" discipline from the community's playbook, and this node is built to make it easy.

One honest caveat: this is the heavyweight of the pack. Every time index changes, the node applies the LoRA to the model - that's a real per-cell cost, and if you're also running a heavy checkpoint it's not free. It's still cheaper than re-loading a checkpoint per comparison.

Installing it

Part of comfy-easy-grids. ComfyUI Manager: search "comfy-easy-grids". Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/shockz0rz/comfy-easy-grids

Then restart. No extra dependencies and no model files - it only needs the LoRAs you already have in ComfyUI/models/loras. It lives under the "EasyGrids" menu with the rest of the pack.

Troubleshooting

If a dropdown is missing a LoRA you know you have, restart ComfyUI - the list is read when the node loads. If the output looks exactly like the base model, check the obvious: index pointing at a "None" slot, or both strengths at 0.0 (which the node treats as "do nothing" on purpose). And remember indexes start at 1 here, so slot 1 is index = 1, not 0.