LoraLoaderModelOnly_motorway_edition
LoRA on the model only — when you don't want to touch the text encoder
- MOTORWAY 🚌💨
- MOTORWAY 🚌💨
The name says it all: LoraLoaderModelOnly_motorway_edition applies a LoRA to the diffusion model and only the model. No CLIP patching, no second strength knob, no text encoder involvement. It pulls the model out of your Motorway pipe, patches it, and writes the patched model back.
Why would you ever want that, when the full LoraLoader does both? Because not every LoRA touches the text encoder meaningfully. Many style and concept LoRAs are trained almost entirely on the model side, and patching CLIP with them does nothing useful - it's wasted compute and, in a few awkward cases, it can muddy the prompt embedding. Model-only loading is also the standard move in a few specific spots: you're using a LoRA that was explicitly released model-only, you're on a workflow that keeps its CLIP untouchable, or you're applying a LoRA mid-pipeline where only the model matters. It's the lighter, faster, more predictable option, and a lot of power users default to it for exactly that reason.
In a Motorway context this is also the cheaper node to chain: it only needs the model key, not a clip key, so you can stack several of these without worrying about whether CLIP is even in the pipe yet.
The wrapper
MOTORWAY 🚌💨- the pipe. Required.INPUT_model_key(defaultmodel) - where the base model lives.lora_name- dropdown of LoRA files frommodels/loras.strength_model(FLOAT, -100 to 100, default 1) - how strongly to patch. Negative allowed.OUTPUT_MODEL_key(defaultMODEL) - where the patched model lands.
Output port is the pipe; pull MODEL back out with a Motorway AxB ramp to hand to a plain node, or chain another loader that reads MODEL.
Installing
Part of the Motorway pack:
cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway
# restart ComfyUI
Or ComfyUI Manager → search ComfyUI_agilly1989_motorway → Install. No pack-specific downloads; LoRAs live in your models/loras folder.
Gotchas
- The model must already be in the pipe -
Key: 'model' doesn't exist in motorwaymeans the checkpoint never got written upstream. - Keys are exact strings; rename an output key and every reader must follow.
- Some LoRAs genuinely need the CLIP side (character work often does). If you apply a character LoRA model-only and the likeness just isn't landing, try the full
LoraLoader_motorway_editioninstead - that's a data point, not a bug. - Beta pack: author's README says "IF THINGS BREAK ITS BECAUSE I BROKE IT," and generated clones have been toggled off in some releases. Missing node after install = beta/version, not you.
Reach for this when you know the LoRA is model-side or when you want the fastest possible patch. Reach for the full loader when the prompt side matters.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| MOTORWAY 🚌💨 | MOTORWAY 🚌💨 | — | |
| INPUT_model_key | STRING | model | — |
| lora_name | COMBO | 0 options: | |
| strength_model | FLOAT | 1.00-100–100 | — |
| OUTPUT_MODEL_key | STRING | MODEL | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MOTORWAY 🚌💨 | MOTORWAY 🚌💨 | — |