JLC LoRA Loader - Multi Model

Stacking three or four LoRAs in ComfyUI usually means three or four LoraLoader nodes in a row, each one feeding its patched model into the next, with a separate strength slider per node scattered across your canvas. JLC LoRA Loader - Multi Model (class JLC_DynamicLoraLoaderModelOnly, from Damkohler's jlc-comfyui-nodes) replaces that with a single node that holds up to ten LoRA slots and returns one patched model.

The "Model Only" part is the design philosophy, not a limitation. This loader patches the diffusion model and only the diffusion model - it deliberately skips CLIP/text-encoder patching and filters out any text-encoder keys a LoRA file contains. For modern architectures that's often exactly what you want: since the Flux generation of models, most LoRA training doesn't even touch the text encoder, and a model-only stack keeps your workflow simpler with a single MODEL output and no misleading text-encoder warnings. If you do need text-encoder influence, the same pack has MODEL+CLIP variants - this is the plain default choice for a model-side stack.

How it works

All ten slots exist in the workflow behind the scenes. The frontend hides rows above slot_count, and the backend treats that number as authoritative - it ignores everything beyond it. Hidden slot values stay serialized in your workflow JSON, so dropping slot_count from 5 to 2 and back again doesn't destroy your selections. Active LoRAs are applied sequentially in visible slot order, left to right.

The inputs that matter

• model - your base model (or the output of a previous loader).
• slot_count - how many rows are visible and active, 1–10. Default is 1, so the first thing you'll do is crank it up.
• lora_XX - the standard dropdown of everything in your models/loras folder.
• strength_XX - per-slot MODEL strength, default 0.0, ranging -10 to 10. Negative strengths are a real thing here: they let you push against a LoRA's effect.

Output is a single MODEL, straight into your sampler or VAE decode.

Installing it

ComfyUI Manager → search jlc-comfyui-nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Damkohler/jlc-comfyui-nodes.git

Restart ComfyUI and you'll find it under loaders. No extra dependencies and no model downloads - it just reads whatever is already in your loras folder.

Where people get burned

The number-one trap is the default strength of 0.0. A beginner drops a LoRA into a slot, sees the strength field already filled in, and wonders why nothing changes - because 0.0 is "no effect," by design, so the node behaves sanely when a slot is only half-configured. Turn the strength up and it kicks in.

Second: remember it's model-only. If a LoRA was trained to change how the text encoder interprets your prompt (common with some character/style LoRAs from the SDXL era), this loader will apply the model-side effect and quietly ignore the text-encoder half. That's the intended split - if you notice a LoRA feels "weaker" here than in a CLIP-patching loader, that's the reason, and it's the pack's MODEL+CLIP loader you're after.