LoraLoaderText

The stock LoraLoader gives you a dropdown of every LoRA in your models/loras folder. LoraLoaderText gives you a text field instead. Same load, same outputs - MODEL and CLIP - but the filename is whatever string you feed it, which means you can feed it anything: a wildcard expansion, a line from a CSV, a value computed by another node.

That's the whole trick, and it unlocks the two things people actually build with this pack. One: randomize LoRA selection by wiring a TextWildcards node (or any text source) into lora_name - every generation rolls a different style LoRA. Two: iterate through a fixed list of LoRAs across a batch, so you can A/B a character LoRA against a style LoRA on the same prompt. This was one of the first ComfyUI nodes to make LoRA iteration programmatic, and in 2023 people genuinely used it to build exactly that loop.

Inputs and outputs

• model (MODEL) and clip (CLIP) - passthroughs from your checkpoint loader, returned alongside the patched versions.
• lora_name (STRING, single-line) - the LoRA filename as free text. Type it directly or wire it from another node. It resolves against models/loras and even tolerates missing extensions, plus filename wildcards.
• strength_model and strength_clip (FLOAT, default 1.0) - the usual LoRA strengths, 0–10 in 0.01 steps. Both at 0 short-circuits to a passthrough (no load at all), and an empty or unresolvable name safely returns the model and clip unchanged rather than erroring.

Outputs: patched MODEL and CLIP. If the load fails for any reason, the node swallows the exception and passes the inputs through - a design choice that keeps workflows running but means a bad filename can silently produce an unpatched model. Worth knowing when you're debugging why a style isn't showing up.

Install

Clone and restart; ComfyUI Manager finds it by searching "Lilly":

cd ComfyUI/custom_nodes
git clone https://github.com/lilly1987/ComfyUI_node_Lilly

Restart after cloning. No models to download; the pack auto-installs rich and chardet on first load.

Fair warning: this pack is unmaintained, and the author's README now recommends Impact Pack. The text-driven LoRA-loading pattern itself is still exactly how people build randomized and batched LoRA workflows today - this specific implementation just hasn't been touched since 2023. It works; just don't expect fixes.