Checkpoint Arg (watdafox)

A dropdown of every checkpoint in your models/checkpoints folder, with two outputs: the selection as a value and the same selection as a plain string. Checkpoint Arg doesn't load anything - no model hits VRAM, no weights move. It's the "which model do I want" decision, decoupled from the actual loading.

That decoupling is the point. In a stock workflow your checkpoint is chosen inside the Checkpoint Loader, buried in the middle of the graph. Here's the scenario that makes this node earn its keep: you've got a workflow you share or re-run across models, you want to flip between dreamshaperXL and realvis without hunting through the graph, and you want the model name embedded in your output metadata so you can tell what made what later. This node gives you a single visible dropdown plus a str_ckpt_name output that will happily flow into any text/logging node.

How it works

The combo is populated from folder_paths.get_filename_list("checkpoints"), so it lists whatever .safetensors (or .ckpt) files you have in the checkpoints folder - same source the stock loader uses, so you'll never see an entry that fails to load. The execute returns (ckpt_name, ckpt_name): the first output is the enum value that slots into a loader's ckpt_name input, the second is the identical string.

Here's the workflow trick: wire the ckpt_name output into CheckpointLoaderWithOuputDirByModelName (this same pack) or the stock Checkpoint Loader, and you can swap models from one spot. Since this node is a pure passthrough with no state, changing it is cheap - nothing reloads until the loader actually runs.

What you touch

• ckpt_name - the combo. Pick a model.
• Outputs ckpt_name (into a loader) and str_ckpt_name (STRING, for logging / text nodes).

Menu location: watdafox/parameter. The pack's Unet Arg, VAE Arg, sampler and scheduler Arg nodes all follow this same "pick once, reuse, log always" pattern, so if you like one you'll like the family.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/OhSeongHyeon/comfyui-watdafox-nodes.git

Restart ComfyUI, or use ComfyUI Manager and search "comfyui-watdafox-nodes". No pip dependencies, no model downloads - this node just reads your existing checkpoints folder.

When it's actually useful

If your workflow is a single checkpoint and you never log metadata, this is redundant ceremony. It shines when you A/B test models, maintain workflow templates that get re-pointed at different checkpoints, or want the model name in your saved info. If you're doing the last one, remember: use str_ckpt_name, not the enum output, when feeding text nodes.