Checkpoint Loader 🐍

Same job as ComfyUI's built-in checkpoint loader - read a full model file off disk, hand back the pieces everything else needs - wrapped in the same picker upgrades this pack gives its LoRA loader: preview thumbnails, saved sample prompts, and a proper info panel, instead of a bare filename dropdown.

It's from ComfyUI-Custom-Scripts, pythongosssss's utility pack. Once you've got more than a handful of checkpoints on disk, this is the loader that saves you from squinting at filenames trying to remember which one is which.

What a checkpoint actually is

Worth a beat if you're new: a checkpoint is the complete set of weights defining everything a model can generate, unlike a LoRA, which only nudges an existing checkpoint's behavior. Load one, type a prompt, get an image - a checkpoint is self-sufficient. That's why this node's outputs are the foundation everything else in a typical workflow builds on top of.

The inputs and outputs that matter

ckpt_name is a dropdown populated from your models/checkpoints folder - that's the one thing you actually set. There's also an optional prompt text input, mirroring the pack's LoRA loader, mostly relevant for scripted/automated setups rather than everyday manual use.

Outputs are MODEL, CLIP, and VAE - wire them to your sampler, text encoder, and VAE decode exactly like you would with the stock node. There's a fourth output, example, which surfaces a saved sample prompt for that checkpoint if one exists on disk for it.

The quality-of-life layer

Drop a .png or .jpg with the same base filename next to your checkpoint (sdxl.safetensors + sdxl.png) and it shows on hover in the picker. Right-click any output image and choose "Save as Preview" to generate one straight from a result you just got, rather than sourcing one separately. The README also mentions optional subfolder support (toggled in settings) if your checkpoints are organized into folders rather than sitting flat in models/checkpoints. And like the LoRA loader, right-clicking a checkpoint in the picker gives you "View Info" for its metadata without leaving the app.

Installing it

• ComfyUI Manager - search "ComfyUI-Custom-Scripts", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/pythongosssss/ComfyUI-Custom-Scripts, restart.

No models ship with the pack - you point it at checkpoints you already have in models/checkpoints.

Common issues & troubleshooting

Dropdown is empty. Nothing in models/checkpoints, or ComfyUI hasn't been restarted/refreshed since you added a file.

No thumbnail on hover. The preview image needs the exact same base filename as the checkpoint file and needs to live in the same folder - a typo or mismatch means silently no thumbnail, not an error.

No sample prompt in the example output. Same story as the LoRA loader: this needs a matching sidecar file convention, and it's opt-in. Not every checkpoint has saved examples; you build that up yourself as you go.

Checkpoints in subfolders aren't showing up. Subfolder support is a setting, not default-on. Check the pack's settings menu if your models/checkpoints uses nested directories.