Simple Load Checkpoint

Loading a checkpoint is the most over-taught step in ComfyUI, and yet the stock loader still annoys you in two specific ways if you came from WebUI: there's no CLIP skip next to the model, and swapping the VAE means adding a separate node and rewiring. SimpleSyrup.SimpleLoadCheckpoint bundles all of it into one loader that outputs the same model, clip, and vae trio you'd get anywhere else - just with the two knobs you actually reach for sitting where you'd look for them.

The pack's author has said the whole thing grew out of moving A1111/WebUI workflows into ComfyUI, and this node is that migration in miniature. In WebUI, CLIP skip sits beside your checkpoint as a per-model setting, not as a mystery node somewhere off-screen. Here it's a checkbox on the loader.

Inputs, and they're all you need:

• ckpt_name - the checkpoint file. This supplies the base MODEL, CLIP, and the checkpoint's own VAE.
• vae_name - defaults to Use Checkpoint VAE. Change it to swap in any VAE you have in your vae folder. This is the fix for the classic "model loads fine but colors look washed out / artifacts everywhere" complaint, where the model was trained against a different VAE than the one bundled. One dropdown instead of a separate VAE loader node and two extra wires.
• clip_skip - a boolean, default off. Flip it on and prompt encoding uses ComfyUI's CLIP-skip behavior, i.e. condition on the next-to-last CLIP layer instead of the last.

On CLIP skip, a word of context so you don't treat it as a free quality dial. It's a convention that comes from the Danbooru-anime lineage: NovelAI conditioned on CLIP's penultimate layer back in 2022, and a huge share of anime checkpoints (through Pony, Illustrious, NoobAI) inherited that expectation. If your checkpoint's release post says "Clip Skip: 2," flip this on. If you're on SDXL or anything with an LLM text encoder, leave it off - the setting doesn't even exist on Qwen/T5-encoder models, and on SDXL setting it low is how you turn Pony into noise. So: this checkbox is load-bearing for exactly one model family, and inert for the rest. That's fine - it's the same behavior the stock ComfyUI CLIP Set Last Layer node gives you, just baked into the loader so you can't forget it.

Outputs are the standard model, clip, vae. Nothing exotic downstream - wire them to your KSampler and text encoders exactly like any other loader, and if you swapped the VAE, the vae output is the replacement, not the checkpoint's original.

Install: the node ships in SimpleSyrup:

cd ComfyUI/custom_nodes
git clone https://github.com/Artificial-Sweetener/SimpleSyrup
cd SimpleSyrup && pip install -r requirements.txt

or ComfyUI Manager → search SimpleSyrup → Install → restart. SimpleSyrup runs on ComfyUI's v3 extension API, so keep ComfyUI updated. No extra models ship with the loader - it reads whatever's in your checkpoints and vae folders. If a checkpoint you know is in the folder doesn't show up, that's a folder-path problem on your end, not the node's.