Lora Selector

LoRAs are the dominant way to add a character, style, or concept to a checkpoint - small adapter files that patch the model instead of replacing it. In vanilla ComfyUI you load them with a LoraLoader that both names the file and applies it, right there in the sampling path. IPT-LoraSelector does something slightly different: it selects a LoRA and hands you a lora_stack entry for it, so the choice becomes data you can carry through image_info, store in saved metadata, and vary in an XY plot - while the actual application happens in a dedicated loader (Use Loaded Model or the pack's Lora Stack Lorader) that knows what to do with a stack.

If you've ever rebuilt a workflow to change one LoRA and had to find every node that referenced it, this is the pattern that ends that chore. It's the same philosophy as rgthree's Power Lora Loader but aimed at the metadata-centric workflow: LoRA identity travels with the rest of the recipe.

How it works

You pick a LoRA from a dropdown populated from your models/loras folder, set a strength, and the node produces a single-item lora_stack. Connect an existing lora_stack input and it appends, so you can chain selectors to build a multi-LoRA stack. The outputs include lora_name and strength as plain values, which is handy when you want to echo what was used into a filename or a caption without digging through the stack object.

There's a genuinely useful right-click action: View Model Info... looks up the selected LoRA on Civitai using its SHA256 (a cached sha256 input exists for the fallback path) and can show the LoRA's tag list with frequencies. That's the "what is this LoRA actually for" answer at the moment you're staring at a dropdown.

Inputs and outputs that matter

• lora (required, COMBO) - pick the file.
• strength (FLOAT, default 1.0) - application weight; 0.5–0.8 is commonly the sweet spot, and you can go negative for negative-LoRA tricks.
• lora_stack (optional input, then output) - chain in an existing stack to append, or take the single-item stack out.

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/kinorax/comfyui-info-prompt-toolkit.git
cd comfyui-info-prompt-toolkit
pip install -r requirements.txt

Restart, or ComfyUI Manager → "ComfyUI-Info-Prompt-Toolkit". ComfyUI 0.17.0+. No models to download for the node itself; it just reads whatever LoRAs you already have.

Common issues

The classic misunderstanding: this node does not apply the LoRA to a model. Its output is a stack, and something downstream has to consume it - Use Loaded Model does this internally by default. Wire the stack nowhere and you'll wonder why nothing changed. Also, the strength range is wide (−100 to 100) because negative weights are legitimate, but a casual strength=2.0 will often wreck the image; treat values above ~1.5 as deliberate experimentation. The selector reads your loras folder at node-build time, so if you drop in a new LoRA and it doesn't show in the dropdown, refresh the node (or restart) - the list isn't live-polled.