DIGIT LoRA Loader

Confusing naming alert, and it's worth sixty seconds of your time: this pack ships two nodes that both display as "DIGIT LoRA Loader". The one you want for actually applying a LoRA to your model is DigitLoraLoader (the one that outputs model, clip, and trigger_words). This one - DigitLoRALoader - is the training-suite companion. It doesn't touch weights at all. It resolves the path to a trained LoRA and reads the trigger metadata out of the digit_metadata.json your training run wrote, then hands you the strings.

So when is it useful? When you've trained a LoRA with the DIGIT trainer and you're building the inference workflow that uses it. This node gives you the lora_path, the exact trigger_word, trigger_class, and trigger_phrase from the training run - the metadata that would otherwise be scattered across a config file - and lets you wire those strings into a Prompt Combine. Then you apply the actual weights with the other loader (or any stock loader).

How it works

It takes the lora_name (dropdown from your loras folder) or a lora_path_override, resolves it to a path, and looks for a digit_metadata.json in the same directory. That's the file the DIGIT LoRA Trainer writes when a run finishes, carrying the trigger info and training settings. If it finds it, you get the trigger fields and the full metadata as JSON; if not (say, an external LoRA you didn't train here), it still returns the path and a note that no DIGIT metadata exists.

The inputs:

• lora_name - select a LoRA from the ComfyUI loras folder.
• lora_path_override - absolute path to a DIGIT LoRA directory or .safetensors file. Takes priority over lora_name.
• strength - 1.0 default, −2.0 to 2.0. Stored in the metadata output for the record; this node doesn't apply it to anything.

Outputs:

• lora_path - resolved full path.
• trigger_word - the trained trigger.
• trigger_class - the class noun (e.g. "man" or "style").
• trigger_phrase - the full phrase if the run defined one.
• metadata - the JSON blob from digit_metadata.json (with the applied strength added).

Install

cd ComfyUI/custom_nodes
git clone https://github.com/thedepartmentofexternalservices/comfyui-digit.git
cd comfyui-digit
pip install -r requirements.txt

(Or ComfyUI Manager → search comfyui-digit → install.) Restart ComfyUI, look under DIGIT.

Common issues

The realistic failure is a missing digit_metadata.json. If the node returns empty trigger fields, the file isn't next to the LoRA - either it's an external LoRA (expected, and the note in metadata tells you so) or the training run didn't write it. There's nothing to debug in the node itself; it's a file lookup.

The bigger trap is picking the wrong "DIGIT LoRA Loader" off the menu. If your model and clip aren't getting modified, you grabbed this one - the training companion - instead of DigitLoraLoader. The display names are identical, so check the outputs on the node before you wire it. It's a real footgun in an otherwise tidy pack, and the only thing keeping it honest is that this node's outputs are all strings.