SDNQ LoRA Loader

Every modular workflow needs a LoRA step, and this is the one for the comfyui-sdnq-splited pack. The difference from ComfyUI's stock LoraLoader is structural: the stock node patches a ModelPatcher, while this one works on a diffusers DiffusionPipeline - because that's what SDNQ Model Loader hands you. Same job, different plumbing underneath. It applies LoRAs using diffusers' standard API (load_lora_weights + set_adapters), which is exactly what the quantized SDNQ model expects.

Where this node stands out is the 10 LoRA slots. You get lora_name_1 through lora_name_10, each with a matching lora_wt_1 through lora_wt_10 strength (default 1.0 = full strength). Ten is more than anyone genuinely uses in one pass - the UI hides unused slots via JS, so don't feel obligated to fill them - but it's handy for stacking a style LoRA with a character LoRA and a detail one without daisy-chaining loader nodes. Leave unused slots on None.

The inputs that matter

• model - the MODEL output from SDNQ Model Loader. This is the only required input.
• lora_name_1..10 - a dropdown populated from your ComfyUI loras folder (via the same folder scan the stock loader uses). Select a file by name.
• lora_wt_1..10 - strength per slot. 1.0 is full effect; values above exaggerate, negatives invert (useful for suppressing a baked-in style). The README documents the range as −5.0 to +5.0.

Output is model again, wired into the sampler of your choice.

One context worth knowing: LoRAs for FLUX.2 specifically are rare, because training them needs a serious GPU - the KB notes people cite needing something like an RTX 6000 PRO and many hours, which is why the Flux2 LoRA ecosystem stayed thin. For the other SDNQ-catalog models (FLUX.1, SDXL-class, Qwen-Image) the usual LoRA wealth applies. Don't expect a big library of Flux2 LoRAs to appear in that dropdown.

Install

Standard for this pack:

cd ComfyUI/custom_nodes/
git clone https://github.com/ussoewwin/comfyui-sdnq-splited.git
cd comfyui-sdnq-splited
pip install -r requirements.txt

Restart ComfyUI and it shows up under loaders/SDNQ. Manager install is fine but you may need to lower the Security Level (or use "Install via Git URL") to get past the default-channel restriction.

Gotchas

The tooltip on strength calls 1.0 "full strength," but diffusers adapters can interact when stacked - if two LoRAs fight each other, drop the weaker one's weight rather than the stronger. And a real edge case from the source: if the node can't reach your loras folder (a broken ComfyUI import), it silently degrades to just None in the dropdown - if the dropdown is empty, that's a ComfyUI environment problem, not a LoRA-format problem. Finally, remember the README's standing caveat applies here too: this pack is tested on FLUX.2, so LoRA behavior on other catalog models is best-effort.