Lora Extractor

A quick honesty note up front: this node isn't covered in the pack's README, and its schema is about as bare as it gets - one multiline string in, one STRING out. What follows is the most reasonable read of it given the name, the category it lives in, and how the rest of the pack talks about text-form loras - not a line-by-line confirmed spec. If you're relying on exact behavior, test it on a throwaway prompt first.

The name and shape strongly suggest it does what it says: takes a chunk of prompt text and pulls the a1111-style <lora:name:weight> tags out of it, handing back just the matches (or the cleaned remainder - the schema alone doesn't distinguish which side of the split you get). That syntax is the one A1111-era users are used to writing directly in their prompt, and it's the same format the JNodes ImageDrawer's model cards can copy for you (see its "copy the lora in a1111 text form" button). ComfyUI's native lora loader doesn't understand that syntax at all - it wants a dropdown pick and a weight slider, not inline text - so a node that isolates those tags exists to bridge the gap: either for display, for logging what loras a prompt actually calls for, or as a feed into a separate loader like Coziness (which the JNodes README itself recommends for turning text-form loras into actual loader nodes).

How it works

Feed it a multiline string - realistically, a full prompt that may or may not contain one or more <lora:...> tags mixed in with your normal descriptive text. It returns a single STRING. Given the node's place among JNodes' other prompt-cleanup nodes (SearchAndReplace, RemoveParseableDataForInference), the likely job is scanning the input for that bracket syntax and returning what it finds, separate from the rest of the prompt.

The inputs and outputs

• string (STRING, multiline) - the text to scan, typically a full prompt.
• STRING output - the extracted result.

How to install it

Via ComfyUI Manager: Install Custom Nodes, search "JNodes", install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/JaredTherriault/ComfyUI-JNodes
pip install -r ComfyUI-JNodes/requirements.txt

then restart ComfyUI. No model downloads - it's plain text processing.

Common issues & troubleshooting

Verify the output on a known input before trusting it in a real workflow. Because the exact split (matches vs. remainder) isn't documented, the safest first move is to wire the output straight into a Show Text-style node and run it against a prompt with one obvious <lora:test:1.0> tag in it, so you can see with your own eyes which half you're getting.

It won't load the lora for you. Whatever this node extracts is still just text - you still need a real loader node (ComfyUI's native one, or a text-driven loader like Coziness) downstream to actually apply the lora to your model and CLIP. Don't expect a MODEL or CLIP output here; there isn't one.

Trigger words aren't loras, and this node is about the <lora:...> tag, not trigger words in your text. If a lora needs a specific trigger word in the prompt to activate its effect, that word lives in your normal prompt text, not inside the lora tag - this node has no way to know what a given lora's trigger word is, since that's metadata that lives on the model file itself, not the prompt.