AC_Super_CKPT&LCM

AC_Super_CKPT&LCM collapses two of the most common loader steps into one node: it loads a checkpoint and applies a LoRA in a single pass, handing you MODEL, CLIP, and VAE at the end. If you're tired of dragging a wire from Load Checkpoint into a LoraLoader and then back out again, this is the compression you want.

What it's really doing

The node name is about the workflow it's aimed at, not a special mechanism. In the source it's a plain load_checkpoint_guess_config followed by load_lora_for_models - the exact same operations the core Load Checkpoint + LoraLoader chain performs. The "LCM" part comes from this pack's house style: the author (Cc啊程, a ComfyUI teacher on Bilibili) builds around fast LCM sampling, where you load a regular checkpoint plus an LCM LoRA and then sample at a handful of steps. The node itself doesn't care; it's just checkpoint-then-LoRA.

The inputs that matter

• ckpt_name - your checkpoint, from models/checkpoints.
• lora_name - one LoRA, from models/loras.
• strength_model and strength_clip - both default to 1.0. If both are 0, the node skips the LoRA entirely and just returns the loaded model - which makes it a checkpoint loader with an ignored LoRA slot. Strengths range from −10 to 10, so you can also use a negative value to actively subtract a LoRA's influence, a trick people use for "anti-LoRA" style removal.

The outputs

MODEL, CLIP, and VAE, wired exactly like a Load Checkpoint node's outputs. MODEL and CLIP have the LoRA baked in; VAE is untouched. From here you'd feed CLIP into a text encoder and MODEL into a sampler - or into the pack's AC_Super_KSampler, which takes the clip and does the rest for you.

Installing it

It's part of ComfyUI_AC_FUNV8Beta1. ComfyUI Manager (search "AC_FUN") or:

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

Restart ComfyUI. Nothing to download - it reads your existing checkpoint and LoRA folders.

Where people get burned

The obvious trap is expecting more than it does: one checkpoint, one LoRA, done. If you need a second LoRA stacked, this isn't the node - that's what AC_Super_Lora&LCM's Merge mode is for. And since it's a single fused operation, you can't inspect the intermediate model before the LoRA applies. If you're debugging whether a LoRA is the thing wrecking your output, keep the two core nodes separate. This one is for when you've got a known-good combo and just want it shorter.