Simple Extract Lora From Text 📚

If you've ever copied a prompt off CivitAI or out of a Forge/A1111 generation and dropped it straight into a ComfyUI text box, you've hit this: the prompt has <lora:detailed_skin:0.8> sitting right in the middle of it, and ComfyUI has no idea what to do with that - CLIP Text Encode just sends it through as literal text, which does nothing useful and can actively confuse the model. Simple Extract Lora From Text exists to solve exactly that mismatch: it splits a mixed string into "the LoRA tags" and "the actual prompt," so each half can go where it's supposed to.

How it works

It's a text node, nothing more - feed it one string, and it scans for the Forge-style LoRA syntax: <lora:name:1.0> or the more detailed <lora:name:unet=1.0:te=0.75> form that splits UNet and text-encoder strength separately. Everything matching that pattern gets pulled into lora_text; everything else - your actual subject, style, and quality tags - comes out clean as prompt. No LoRA files get touched here, no model gets modified; this node is purely a text splitter. The actual application happens downstream in Simple Lora Loader, which is the node this one is built to feed.

The inputs and outputs that matter

There's exactly one input, text - the mixed string with LoRA tags embedded in it - and two outputs, lora_text and prompt. Wire prompt into your CLIP Text Encode, and wire lora_text into a Simple Lora Loader node, which will parse those tags and actually apply the LoRAs to your model and clip.

Installing it

Via ComfyUI Manager: search ComfyUI_SimpleButcher, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/KLL535/ComfyUI_SimpleButcher.git
cd ComfyUI_SimpleButcher
pip install -r requirements.txt

then restart. This is a pure text-parsing node - no model downloads, no meaningful dependency weight.

Common issues

Only Forge-style tags get recognized. If your source text uses a different LoRA notation - some tools write things slightly differently, or you've hand-typed a variant syntax - this node won't catch it and it'll pass straight through as part of prompt instead of being extracted. Stick to the <lora:name:weight> or <lora:name:unet=X:te=Y> forms and it'll behave.

It's not doing anything by itself. This node has no visible effect on your generation on its own - it just reorganizes text. If you wire it up and nothing seems different, that's expected; the payoff shows up once lora_text reaches a Simple Lora Loader and the LoRAs actually get applied to the model.

Where this earns its keep is batch workflows. The natural use is with Simple Load Line From Text File upstream: write a prompt file where some lines have LoRA tags baked in and some don't, feed each line through this node, and you get automatic per-line LoRA switching without hand-editing a separate LoRA loader for every prompt. That's the whole reason the pack's author built it - Forge lets you write LoRAs inline in the prompt text itself, and this is the piece that makes that habit portable into ComfyUI.

It's a small, single-purpose node - there isn't much surface area for things to go wrong, which is honestly the appeal. If your prompt has no LoRA tags in it at all, lora_text will simply come back empty and prompt will pass through unchanged, which is the correct, boring behavior.