Nodes/LoRA Txt Loader/LoRA Txt Loader (From Path)
ComfyUI Node

LoRA Txt Loader (From Path)

The LoRA loader that reads the .txt itself, for path-driven workflows

By kwokkakiu233·Created 2 months ago·Updated about a month ago· 2
LoRA Txt Loader (From Path)
  • model
  • clip
  • model
  • clip
  • positive_prompt
  • lora_path
lora_path
strength_model1.00
strength_clip1.00

LoRA Txt Loader (From Path) is the variant of this pack for people who don't want a dropdown at all. Instead of picking a LoRA from a list, you hand the node an absolute file path string, and it loads the model and reads the trigger-word .txt itself. No browser, no frontend auto-fill - the file reading happens in Python at execution time.

That makes it the right node for dynamic workflows. If another node computes or passes along a LoRA path, or you're driving ComfyUI from its API where no one is around to click a tree, the path input is what you want. It's the only member of the family that's fully self-contained on the trigger-word side: you don't have to wire up a separate text-loader to get words into the graph.

How it works

The inputs are the giveaway. There's no positive_prompt box here - just:

  • model / clip - from your checkpoint loader.
  • lora_path - the full path to a .safetensors (or .ckpt) file.
  • strength_model / strength_clip - standard strengths, default 1.0.

The node checks the file exists, loads it, and applies it via comfy.sd.load_lora_for_models, same as the native Load LoRA. Then it looks for a .txt with the same base name next to the model - models/loras/my_lora.safetensorsmodels/loras/my_lora.txt - and reads it. That content comes out of the positive_prompt STRING output automatically.

Outputs are model, clip, positive_prompt (the auto-read trigger words), and lora_path (echoed back).

The gotcha that gets people: if the path is wrong or the file is missing, the node raises a FileNotFoundError and the queue dies with it. So keep the path absolute and correct - no relative paths, no tilde shortcuts. And remember the words still only matter if you wire the positive_prompt output into a CLIP Text Encode (or Text Concatenate → text encode). Patching model and clip alone applies the LoRA but leaves the trigger out of your prompt.

Installing it

Same pack, same install - ComfyUI Manager search "LoRA Txt Loader", or:

cd ComfyUI/custom_nodes
git clone https://github.com/kwokkakiu233/comfyui-lora-txt-loader

Restart after. No requirements.txt, no extra Python dependencies.

When to actually use it

If you're building normal interactive workflows and just want trigger words auto-filled in a box, use the pack's main LoRA Txt Loader - the tree browser is nicer for that. Reach for From Path when the LoRA is chosen programmatically: a workflow that iterates a list of paths, a node that outputs a path from some other logic, or API-driven batches where the graph needs to run unattended. It's a niche, but it's the niche this node fills cleanly.

Categoryloaders

Inputs (5)

NameTypeDefaultDescription
modelMODEL
clipCLIP
lora_pathSTRING
strength_modelFLOAT1.00-10–10
strength_clipFLOAT1.00-10–10

Outputs (4)

NameTypeDescription
modelMODEL
clipCLIP
positive_promptSTRING
lora_pathSTRING