Checkpoint Selector (Yogurt Nodes)

The normal Checkpoint Loader gives you a model, a CLIP, and a VAE. What it doesn't give you is a string - the name of the checkpoint as plain text you can feed into a save filename, a prompt, a conditional, or an API call. Checkpoint Selector is the version of that loader that exposes the metadata. Pick your checkpoint from a dropdown, and out the other side you get the model itself plus name, stem, trigger_word, and absolute_path as strings.

What it's really for

It's not a loader replacement for everyday generation - the stock loader is fine for that. This node earns its place in workflows that need to know what model they're using. The classic example is naming output files: you want output_<model-stem>_<seed>.png so your results folder is searchable, and that requires the model name as a string somewhere in the graph. Same story for logging, for driving a script, or for the pack's own ComfyUI Client nodes where you're building a workflow JSON and want to inject the checkpoint name dynamically.

The node scans your models/checkpoints folder, so the dropdown lists whatever you actually have installed - the brief's schema snapshot shows the list as ["None"], which is just what an empty snapshot looks like; at runtime it fills with your real files. name is the full filename, stem is the filename without the extension (the one you'll actually use in filenames), and trigger_word is interesting: it's an input and an output. You type a trigger word to remember with this checkpoint, and it comes back out as a string - so you can jam checkpoint_name + trigger_word into your prompt automatically. It's a small thing, but it removes a whole class of "which trigger word was this model again" fumbling.

Inputs

• checkpoint - the dropdown of your installed checkpoints.
• trigger_word - a string you associate with this checkpoint, passed through to the output.

Outputs

• checkpoint - the model itself, ready for a standard sampler chain (the output type is dynamic so it connects to whatever expects a checkpoint).
• name - full filename, e.g. juggernautXL_v9Rundiffusion.safetensors.
• stem - filename without extension: juggernautXL_v9Rundiffusion.
• trigger_word - whatever you typed, echoed back.
• absolute_path - the full path to the file on disk.

Install

Standard Yogurt Nodes install. ComfyUI Manager → search "ComfyUI-YogurtNodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes.git
cd ComfyUI-YogurtNodes && pip install -r requirements.txt

Then restart ComfyUI. No downloads - it just reads your existing checkpoint folder.

Where people get burned

The main trap is expecting it to behave like the regular loader. It hands you the model, yes, but many workflow templates wire a VAE and CLIP out of the checkpoint loader too - this node doesn't give you those, because it's a selector, not a full loader. If your graph needs a separately loaded VAE (and let's be honest, on SDXL you often want a good VAE loaded by hand anyway), you'll wire that in separately. The other thing: because it's a string-driven node, it only knows about files physically present in models/checkpoints. Subfolders work, but if you keep your checkpoints on a NAS or symlink them oddly, the dropdown can look different from what the regular loader sees. When the list looks short, check your folder structure first. It's a utility node with one job - know your model's name and path - and it does that job cleanly.