Lora Loader (JPS)

Functionally this is ComfyUI's stock LoRA loader - takes a model and clip, applies a LoRA at whatever strength you set, hands back a modified model and clip. The one difference that actually matters is the switch widget: flip it to Off and the node passes model/clip straight through untouched, LoRA not applied, no rewiring or bypassing required. Flip it back On and the LoRA's back. That sounds small until you're mid-session trying to figure out whether a particular LoRA is actually helping or just adding noise - with the stock loader you'd delete the node or bypass it and reconnect; here it's one click.

How it works

switch (Off/On) gates whether the LoRA gets applied at all. Per the pack's own README, this widget's underlying value follows the same 1-indexed convention (1/2) the rest of the pack's switches use rather than a plain 0/1 boolean - which is why the README specifically calls out that it "works with most 'x to 1'-switches" without needing extra conversion logic, unlike some alternatives that use boolean values and need corresponding math nodes. You won't notice this day to day (it's just Off/On in the UI), but it's a small sign the whole pack was built with a consistent internal convention rather than each node doing its own thing.

The inputs that matter

• lora_name - a dropdown reading whatever .safetensors files are sitting in ComfyUI/models/loras at startup. This is a live filesystem list, not a fixed set - if a LoRA you just downloaded isn't showing up, it wasn't there when ComfyUI booted.
• strength_model / strength_clip (-10 to 10, default 1, step 0.1) - the usual dual-strength split most LoRA loaders expose, letting you weight the LoRA's effect on the diffusion model and the text encoder separately. Most LoRAs behave sensibly somewhere in the 0–1.5 range; the wide negative range is there if you're deliberately trying to subtract a LoRA's effect, which is a niche move but a real one.
• switch - Off/On, as above.

Outputs: MODEL and CLIP, wired onward exactly like any other LoRA loader - into your KSampler chain and your CLIP Text Encode nodes respectively.

Installing it

ComfyUI Manager: search JPS Custom Nodes for ComfyUI. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/JPS-GER/ComfyUI_JPS-Nodes.git

Restart ComfyUI. Delete a previous install of the pack first if you have one. No models bundled - you need actual LoRA .safetensors files sitting in ComfyUI/models/loras before there's anything in the lora_name dropdown to pick.

Where people get tripped up

If a LoRA doesn't show up in lora_name, it's a file-location problem, not a node problem - check it's actually in models/loras and restart ComfyUI (a running instance won't pick up files dropped in mid-session). And if you're chaining multiple LoRAs, remember this node applies one LoRA per instance - for two or more, chain the MODEL/CLIP outputs of one Lora Loader into the inputs of the next, the same pattern the stock loader uses.