Simple SDXL Loader 📀

A checkpoint, its CLIP text encoder, and its VAE are three separate things that every workflow needs at the same time. ComfyUI's stock answer is the Checkpoint Loader, which hands you all three from one socket trio. The Simple SDXL Loader from TensorVizion/OmniNodes is that same trick with two conveniences bolted on: a VAE auto-detect that fills in a missing decoder, and a summary output that documents what got loaded.

The name oversells it slightly. "SDXL" is the pack author's focus, but underneath it calls ComfyUI's own load_checkpoint_guess_config, which guesses the architecture and works for SD 1.5 and a lot of other checkpoints too. It's a thin, opinionated wrapper - the opinion being "one node should give you the full stack."

Inputs and outputs

Three inputs, none of them mysterious:

• model_name - a dropdown populated from your registered checkpoints/ folder. Only files ComfyUI knows about show up here; if yours is missing, it's in the wrong directory, not a node problem.
• vae_auto (default on) - the interesting one. If the checkpoint has no baked-in VAE, the loader goes looking for a same-named file in your vae/ folder - MyModel.safetensors becomes MyModel.vae.safetensors. It's a thoughtful touch for SDXL community finetunes that ship decoder-less.
• vae_name - an explicit override. Set this to force a specific external VAE no matter what the checkpoint bundles; it wins over auto-detect.

Outputs are model, clip, and vae - wire them straight into a KSampler, a CLIP Text Encode, and a VAE Decode - plus a summary string that reports where the VAE came from (bundled, auto-detected, or overridden).

When you'd bother

Honestly, if you already have stock Checkpoint Loader muscle memory, this node buys you two things: the VAE auto-detection and the summary. The summary is the sleeper feature - when you're trying to figure out why an image looks washed out, "vae_auto found X.vae.safetensors" beats guessing which decoder you loaded. And per the VAE lore, the right VAE is the one the checkpoint was trained against, not the one with the best reputation - so knowing exactly which file got attached actually matters.

Install

It's part of OmniNodes, so install the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/TensorVizion/OmniNodes

Or ComfyUI Manager → search OmniNodes → install → restart. No extra pip packages; it uses ComfyUI's own folder_paths, comfy.sd, and comfy.utils. You'll find it under TensorVizion/Model Utilities.

Troubleshooting

• "Checkpoint not found" - the loader only sees files under ComfyUI's registered checkpoint search paths. Drop the file into ComfyUI/models/checkpoints/ (or the equivalent) and restart.
• VAE stays empty - vae_auto only helps when the checkpoint has no bundled VAE and a same-named .vae.safetensors file exists. Otherwise set vae_name explicitly.
• Node not in the menu - restart ComfyUI and check the terminal for [OmniNodes] load lines; an import error prints a full traceback there.

Reach for this when you're building a workflow inside the OmniNodes universe and want the load side to be one clean node. It's the checkpoint loader you know, with better record-keeping.