Lora Loader Only Model By Path
Load a LoRA straight from disk, no models/loras folder needed
- model
- MODEL
Stock ComfyUI's LoRA loader only sees what's sitting in models/loras. Fine if your whole LoRA collection lives there - annoying if it's scattered across an external drive, a training-run output folder, or a shared network location, and you don't want to copy or symlink every single file just to try one out. This node skips the dropdown entirely and takes a raw file path.
How it works
It's the model-only variant: you feed it a model, a lora_path string pointing at a .safetensors file anywhere on disk, and a strength_model, and it applies the LoRA to the diffusion model weights only - nothing touches CLIP. That distinction matters more now than it used to. Modern architectures (Flux, Z-Image, and friends) have largely moved away from training the text encoder at all, so a lot of current LoRAs genuinely don't have anything to apply to CLIP in the first place - a model-only loader isn't a compromise for those, it's the correct tool.
The inputs and outputs that matter
model- the diffusion model coming in.lora_path- a literal filesystem path. The default value shown isc:\loras\my_lora.safetensors, which tells you this is a raw path field, not a picker - on Linux or macOS use forward slashes, and remember there's no autocomplete or validation like the built-in dropdown node gives you. A typo here won't get caught until you actually queue the prompt.strength_model- ranges from -20 to 20 in steps of 0.01. That's a much wider range than the usual -2 to 2 you'd expect; realistic values are still almost always close to 1, but the extra headroom is there if you're doing something deliberately aggressive.- Output:
MODEL- wire it straight into your sampler, or chain more LoRA loaders after it.
How to install it
Through ComfyUI Manager: search "SP-Nodes," install, restart. Or:
cd ComfyUI/custom_nodes
git clone https://github.com/bananasss00/ComfyUI-SP-Nodes
Restart ComfyUI afterward. No extra models to download - this node only reads whatever LoRA file you point it at.
Common issues
Because the path field is free text, the most common mistake is a small typo or a path that's relative to the wrong working directory - ComfyUI won't complain until you queue, and the error you get back is usually just "file not found" rather than anything pointing at the exact problem. Double-check you're using the right slash direction for your OS, and that the path is either absolute or correctly relative to wherever you actually launched ComfyUI from, not wherever the workflow file happens to live. If you need CLIP updated too - say the LoRA was trained with a text-encoder component, which is rare on recent architectures but still common on older SD1.5/SDXL LoRAs - reach for this node's sibling, LoraLoaderByPath, instead.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| lora_path | STRING | c:\loras\my_lora.safetensors | — |
| strength_model | FLOAT | 1.00-20–20 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |