AC_Super_Lora&LCM

AC_Super_Lora&LCM takes a MODEL and CLIP and applies LoRAs to them in one node, with a Merge/Divide mode that controls whether one or two LoRAs land. It's the LoRA loader compressed, aimed at the pack's usual fast-sampling workflow - the "LCM" is the author's house style again, not a mechanism: a base model plus an LCM LoRA is the classic combo for sampling in a handful of steps.

How it works

The core is load_lora_for_models, the same function the core LoraLoader calls. The boolean input decides the behavior:

• Merge applies LoRA 1, then stacks LoRA 2 on top of the result. Both strengths are honored, so you get a genuinely merged model - the standard way to combine a style LoRA with a character LoRA.
• Divide applies only LoRA 1 and ignores slot 2 entirely. Think of it as "use the first LoRA, don't bother the second."

Whichever way it runs, it returns the patched MODEL and CLIP. The node also caches the loaded LoRA file internally, so re-running with the same LoRA skips the file read.

The inputs

• model / clip - from a checkpoint loader.
• lora_name_1 / lora_name_2 - from your models/loras folder.
• strength_model_1 / strength_clip_1 (and the _2 pair) - defaults 1.0, range −10 to 10. Set both to 0 and that LoRA is skipped. Negative strengths subtract a LoRA's influence, which is how people do "anti-LoRA" adjustments.

The outputs

MODEL and CLIP, both patched, wired straight into a sampler or this pack's AC_Super_KSampler.

Installing it

Part of ComfyUI_AC_FUNV8Beta1 by Cc啊程 (Bilibili ComfyUI teacher; README is Chinese). ComfyUI Manager (search "AC_FUN") or:

cd ComfyUI/custom_nodes
git clone https://github.com/A719689614/ComfyUI_AC_FUNV8Beta1

Restart ComfyUI. No downloads.

Where people get burned

The naming is the trap: Divide doesn't "split" two LoRAs into separate outputs - it just applies LoRA 1 alone. If you expected both on the canvas, read the dropdown before you click. And like any stacked-LoRA setup, order matters: Merge applies slot 1 then slot 2, so swapping the two files changes the result. Keep the order intentional.