💊 CR Load LoRA

Functionally this is ComfyUI's standard LoRA loader with one genuinely useful thing bolted on: an On/Off switch. Instead of the usual dance of unplugging a LoRA node to test your workflow without it (and rewiring model/CLIP back together), you flip switch to Off and the node passes model and CLIP straight through untouched. Small addition, but it's the difference between a five-second toggle and rewiring your graph every time you want to A/B a LoRA.

How it works

You feed it model and clip from your checkpoint loader. Pick a LoRA file from lora_name, set switch to On, and it applies that LoRA at strength_model and strength_clip - two separate weights, because a LoRA's influence on the diffusion model and on the text encoder don't have to match. Set switch to Off and none of that happens; the node just relays your model and CLIP unmodified, with the LoRA selection and strengths still sitting there ready for when you flip it back on.

Inputs and outputs that matter

• model, clip - straight from your checkpoint loader.
• switch - On / Off. The whole reason to reach for this over the stock loader.
• lora_name - pick from whatever's in your models/loras folder.
• strength_model, strength_clip - both -10 to 10, default 1.0 each.
• Outputs: MODEL, CLIP (modified or passed through, depending on the switch), plus show_help.

On weight: default 1.0 is a starting point, not a target. It shows up constantly in LoRA troubleshooting threads that 1.0 is often too strong - 0.5–0.8 is a more typical sweet spot, especially once you start stacking more than one LoRA on the same model.

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, restart. The pack ships no LoRAs itself - drop your .safetensors files into ComfyUI/models/loras so they show up in lora_name.

Common issues

Almost every "my LoRA isn't doing anything" report traces back to one of a short list of causes, and they apply here exactly as they would with the stock loader:

1. Base model mismatch. A LoRA trained on one architecture (SDXL, Flux, Illustrious, whatever) generally won't produce meaningful results on a checkpoint from a different family - sometimes it errors, sometimes it just silently does nothing. Check what the LoRA was trained against before blaming the node.
2. Missing trigger word. Plenty of LoRAs need a specific word in your prompt to activate; check the model's page for it.
3. Weight too high. Pull it down toward 0.5–0.8 if results look overcooked or the image degrades.
4. The switch is Off. Sounds obvious, but it's the first thing to check with this particular node given the whole point of it is an easy toggle.

If the node's missing from the search menu entirely, that's the wider Comfyroll pack failing to load somewhere else - check your ComfyUI console for an import error and redo a clean git clone if you see one; that fixes it faster than debugging pip dependencies one at a time.