PD_lora_loader

ComfyUI's own LoRA Loader is already the node you'll use 99% of the time, so let's be clear about what PD_lora_loader actually is: a thin wrapper around ComfyUI's model-only LoRA loader (LoraLoaderModelOnly), with two extras bolted on - it outputs the LoRA filename and the strength as strings. The model in, LoRA in, modified model out. If you never need the filename as data, this node is strictly redundant and you should keep using the core one.

Its real audience is anyone building logging or routing around LoRA usage: the name and strength outputs let you wire "which LoRA ran and how hard" into a text display, a filename, or a report - the same philosophy as the pack's loaders that expose filenames as strings.

How it works

Under the hood it subclasses LoraLoaderModelOnly and calls its load_lora with strength_clip = 0 - which is exactly what "model-only" means: the LoRA is applied to the diffusion model (how the latents are denoised), not to the CLIP text encoder. The author's own description is the honest one: it's for "modifying the diffusion model, altering the way latents are denoised, such as applying styles." Multiple of these can be chained - stack several in sequence and each modifies the model further.

• MODEL output - the modified model, ready for your sampler.
• name - the LoRA filename as a string.
• strength - the strength formatted to two decimals (1.00), as a string.

The strength output is just for reporting; it doesn't affect anything downstream except whatever node you feed the string into. The dropdown for lora_name reads your models/loras/ folder like the core loader.

The inputs that matter

• model - the base MODEL in.
• lora_name - the LoRA dropdown.
• strength_model - -100 to 100, default 1.0. This is the actual dial; below ~0.5 the effect fades, above 1.0 it starts to overshoot and you get artifacts - the usual LoRA rules from the KB's lora-training doc apply unchanged.

Installing

ComfyUI Manager, search Comfyui_PDuse; or:

cd ComfyUI/custom_nodes
git clone https://github.com/7BEII/Comfyui_PDuse
cd Comfyui_PDuse
pip install -r requirements.txt

Restart. No model downloads - it uses your existing LoRA files.

Gotchas

The big one: no CLIP output. This is model-only, so if you need to also modify the text encoder (common for style LoRAs that ship both halves), use the core LoRA Loader or LoraLoader - this node will silently skip the CLIP half and your prompt styling may not apply. Also, because it's a subclass, it inherits the core loader's caching behavior; a changed lora_name triggers a reload, but chaining several of these means each runs its own load on top of the accumulated model, so keep an eye on memory if you stack a lot. For logging-oriented workflows it's genuinely handy; for everything else, it's the pack proving it can wrap the core node - and nothing more.