AC_Super_Checkpoint

AC_Super_Checkpoint is a small node with a simple job: it lets you keep four checkpoints in one place and flip between them with a dropdown. Instead of editing a Load Checkpoint node every time you want to try a different model, you set four filenames once, then switch with Select. It's the "one source of truth, fan out" pattern applied to model loading - the same thing the primitive and switch nodes do for numbers and text, just for checkpoints.

How it works

Under the hood it's exactly the stock ComfyUI checkpoint loader. There are four inputs - ckpt_name_1 through ckpt_name_4 - each a dropdown of every checkpoint in your models/checkpoints folder, plus Select, which chooses model1, model2, model3, or model4. Whichever you pick, the node calls load_checkpoint_guess_config on that file and returns MODEL, CLIP, and VAE, same as the core Load Checkpoint node. Only the selected file is loaded at runtime, so the other three just need to be valid filenames - they cost you nothing in VRAM until you switch to them.

Why you'd reach for it

The honest use case is A/B testing models or swapping between a base model and a refiner without touching the graph. It's also handy for shared workflows where someone else might want to drop in their own favorite model - one dropdown instead of hunting for the loader node. That's about it. If you only ever use one checkpoint, this node is pure overhead and you should ignore it.

Installing it

Part of the ComfyUI_AC_FUNV8Beta1 pack by Cc啊程 (a ComfyUI teacher on Bilibili - the README is in Chinese). ComfyUI Manager, search "AC_FUN", or:

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

Then restart ComfyUI. No downloads, no dependencies beyond what ComfyUI already has.

Where people get burned

If you rename or delete a checkpoint file, its dropdown slot resets - and if a slot is left on an empty value, selecting that model errors out until you pick a valid file. Also, a subtle one: the four names live on one node, so if you're sharing a workflow, everyone needs all four checkpoints present or the graph complains at load. And like everything in this pack, it reads from the standard models/checkpoints folder - your GGUF or unet-style loaders don't count here. Keep it simple: four slots, one Select, standard outputs.