ADLoraLoader

ADLoraLoader is the plain-vanilla LoRA loader of the asyncdiff pack: one LoRA file in, one AD_LORA wire out, and that wire feeds the pack's SD text-to-image sampler. It's about as simple as a node gets in this pack - a dropdown and a weight - but it sits on top of the same multi-GPU machinery as everything else, so it's worth knowing where it fits and where it doesn't.

Why it exists

Everything in asyncdiff_comfyui runs through diffusers pipelines, so ComfyUI's regular LoraLoader output is useless to it. This node loads a LoRA in a format the pack's pipeline can apply, and the AD_LORA output is what the sampler actually consumes. LoRAs are the small adapter files the community uses to teach a base model a style, a character, or a concept without retraining the whole thing - they patch a small set of weights and are meant to be swapped freely.

Inputs are two:

• lora - dropdown of the LoRA files the pack can see.
• weight - default 1.0, and it can go negative. 1.0 is full-strength; around 0.5–0.7 you're getting a flavor; negative weights subtract a style or a concept, which is a trick people use to suppress something the base model keeps doing.

Output is a single AD_LORA, and here's the wiring catch: it feeds the lora input on ADSDSampler - the text-to-image sampler. The upscale sampler and the video samplers don't take a LoRA input. Don't assume every sampler wants one; check the node before you wire.

The combiner, if you need more than one

This node is strictly one-in, one-out. If you want to stack two, three, or four LoRAs - the "power loader" workflow that rgthree made mainstream - you don't chain these. You use the pack's ADMultiLoraCombiner, which takes up to four AD_LORA inputs and merges them into a single AD_LORA that feeds the sampler's one input. For a single LoRA, skip the combiner and wire this directly.

Install

Same routine as the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/SlackinJack/asyncdiff_comfyui.git
cd asyncdiff_comfyui
./clone_asyncdiff_repo.sh   # required - pulls the upstream AsyncDiff framework

Restart ComfyUI, or install via ComfyUI Manager by searching "asyncdiff_comfyui".

Where people get burned

The same two warnings apply to every node here: this is a multi-GPU pack, so on a single card the sampler will just be slower than ComfyUI's native path, and if the LoRA dropdown is empty the pack isn't finding your files - the README doesn't document the folder it scans, so check the pack directory after install. And remember the support note: the author's stated working path is stable-video-diffusion-img2vid, so the SD text-to-image lane this loader feeds is the less-tested side of the pack. The pieces are all real; the docs just aren't.