Prompt Studio LoRA Loader

Prompt Studio LoRA Loader is a model-path LoRA loader that only exposes files inside one top-level folder of your LoRA directory. You tell it a folder name, and it makes exactly those LoRAs visible to Prompt Studio's sidebar - nothing else. That's the whole trick, and it's a good one: instead of scrolling a dropdown of every LoRA you've ever downloaded, Studio offers you only the handful that make sense for the workflow you're running.

It's a "Prompt Studio" node in the true sense - its main purpose is being the attachment point that makes Studio's LoRA section appear in the sidebar. When the active workflow contains one, Studio shows a LoRA list, lets you stack any number of the visible LoRAs with independent strengths, and injects that ordered stack into the workflow snapshot it queues. Normal ComfyUI queues of the saved workflow stay pass-through unless you supply the stack yourself.

How it works

The mechanism is folder scoping. You set lora_type to the name of a top-level folder under any ComfyUI LoRA directory (say, flux). The node then lists ComfyUI/models/loras/flux/... recursively, case-insensitively - flux, Flux, and FLUX all match. LoRAs outside that folder are invisible to Studio, and any filename starting with _ is reserved and hidden.

Inputs:

• model - the MODEL tensor from your checkpoint loader; the node accepts and returns a MODEL, so it drops in anywhere in the model path (replace a model-only LoRA loader, or sit between the checkpoint loader and the sampler).
• lora_type - the top-level folder name. Leave it empty and, on a plain queue, the node is a pure pass-through.
• lora_stack_json - optional JSON stack of {"name": ..., "strength": ...} entries. Prompt Studio writes this; you can hand-write it on the canvas too. It's validated against the folder, so a name outside lora_type errors out rather than silently no-oping. Per-LoRA strengths, and the node caches loaded weights by file mtime so repeated runs don't re-read disk.

The single output is model (MODEL).

Installing it

It ships with Prompt Studio, so the usual one-liner:

cd ComfyUI/custom_nodes
git clone https://github.com/tiko13/ComfyUI_PromptStudio

Restart ComfyUI, or install "ComfyUI_PromptStudio" through ComfyUI Manager. No pip dependencies. The node doesn't call an LLM server at all - but if you're using it inside Studio for real, you'll want a KoboldCpp/Ollama/llama.cpp backend running for the prompt side of things.

Where people get burned

• No LoRAs show up. Your lora_type doesn't match a real top-level folder name, or your LoRAs are dumped loose in the root of the loras directory. LoRAs need to live inside a named subfolder to be visible.
• Hand-rolled stack rejected. If lora_type is empty and you pass a non-empty lora_stack_json, the node refuses - it won't guess which folder you meant.
• Expecting it to pick a LoRA for you. It doesn't select anything; it just exposes a scope. The selection comes from Studio's UI or from the stack you provide.