JDCN_AnyCheckpointLoader
Load a checkpoint from anywhere on disk, not just your models folder
- MODEL
- CLIP
- VAE
JDCN_AnyCheckpointLoader loads a checkpoint, CLIP, and VAE from a path you type in - any absolute path on your machine, not just the models/checkpoints folder. If you've ever had a model living in a second SSD or a network drive and wished the standard loader's dropdown would just show it, this node is that wish, minus the dropdown.
How it works
Three string inputs, all full paths:
ckpt_path- the checkpoint file itself (.safetensors,.ckpt, or any formatload_checkpointhandles).config_path- leave it at the defaultOptionaland the node guesses the model config from the file. Type a path and it loads with that explicit config instead.embedding_folder- again defaultOptional, which maps to ComfyUI's normal embeddings directory. Set it to a folder if your embeddings live somewhere else.
Under the hood it calls ComfyUI's own load_checkpoint / load_checkpoint_guess_config, so the output is the standard trio: MODEL, CLIP, VAE - wire them exactly like you would the built-in CheckpointLoaderSimple. VAE included, no separate load step.
Why you'd use it
Mostly: checkpoints that sit outside ComfyUI's model directory. An organizer keeps ckpt files in one big external library; a shared machine stores models on a mounted drive; you're testing a file before moving it into place. This node skips the copy-to-models-folder step entirely. It also pairs cleanly with the pack's file-list nodes - AnyFileList scans your external folder for .safetensors, AnyFileSelector picks one, and its path feeds straight into ckpt_path.
The flip side, and it's a real one: no dropdown, no validation. Type the path wrong and you get an error; the node doesn't browse or suggest. And you're bypassing ComfyUI's model-folder conventions entirely, so workflows that use this node aren't portable - someone else's machine won't have your D:\models\... path.
A word on security
Since this loads arbitrary paths, keep the same caution you'd apply to any custom node: only load checkpoints you actually trust. A checkpoint is executable code in all but name, and ComfyUI's ecosystem has had malware incidents (the ComfyUI_LLMVISION case) where exactly this kind of unguarded trust was the vector. Loading from a folder you control is fine; don't point it at random downloads.
Installing it
Part of ComfyUI-JDCN:
- ComfyUI Manager β Install Custom Node β search JDCN β install ComfyUI-JDCN β restart.
- Or:
cd ComfyUI/custom_nodes
git clone https://github.com/daxcay/ComfyUI-JDCN.git
cd ComfyUI-JDCN
pip install -r requirements.txt
Restart; it's under π΅ JDCN π΅. The pack only requires piexif; no models to download (this node loads models, it doesn't fetch them).
Common issues
Typos in ckpt_path are the #1 error source - double-check absolute paths, and remember Windows backslashes need care. If config is set to something other than Optional and it's wrong, loading fails; leave it Optional unless you know you need an explicit config. And if the model loads but looks off, check the VAE - the node loads the checkpoint's bundled VAE, and with modern checkpoints that's usually right, but the "swap in a different VAE" workflow still applies when you need a better decoder (see the VAE notes in ComfyUI docs for the one your checkpoint was trained against).
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| ckpt_path | STRING | undefined | β |
| config_path | STRING | Optional | β |
| embedding_folder | STRING | Optional | β |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | β |
| CLIP | CLIP | β |
| VAE | VAE | β |