Nodes/ComfyUI Model Batch Downloader/Load Checkpoint (Downloaded)
ComfyUI Node

Load Checkpoint (Downloaded)

Load the checkpoint you just batch-downloaded, without the folder dance

By watarika·Created 2 months ago·Updated 2 months ago· 0
Load Checkpoint (Downloaded)
  • download_result
  • MODEL
  • CLIP
  • VAE
id

Every standard ComfyUI loader assumes one thing: the file already exists in the right folder. Load Checkpoint (Downloaded) is the version that assumes you just downloaded it - with the Model Download Batch node sitting upstream in the same workflow. It takes the batch's download_result, pulls out the checkpoint you named, and loads it exactly like ComfyUI's own Load Checkpoint does, because under the hood it literally calls the core CheckpointLoaderSimple.

Where it fits

A checkpoint - the "all-in-one" style model, the kind most of the Civitai ecosystem grew up on - bundles three things in a single .safetensors: the diffusion model (the thing that does the denoising), the text encoder (the thing that reads your prompt), and the VAE (the thing that turns latents into pixels). That's why this loader has three outputs. Modern split-style setups like Anima or Krea 2 break those into separate files and use this pack's Diffusion Model / CLIP / VAE loaders instead; this one is for when one file is all you need.

The README's flagship example is Illustrious-family: download a single all-in-one checkpoint into the checkpoints category, then load it with this node. One manifest entry, one loader, done.

Inputs and outputs

Two inputs, both worth understanding:

  • download_result - connect this to the download_result output of either download node.
  • id - a string matching the manifest id you gave that checkpoint. If you left id blank in the manifest, it defaults to the filename without the extension, so mycheckpoint.safetensors is addressable as mycheckpoint.

Outputs are the classic three:

  • MODEL - wire this into the model input of your sampler (KSampler or whatever your workflow uses).
  • CLIP - wire into the two CLIP Text Encode nodes for your positive and negative prompts.
  • VAE - wire into VAE Decode at the end of the pipeline.

If that wiring looks identical to a normal checkpoint loader, that's the point: this node is a drop-in that just happens to resolve the file through the download result instead of a folder dropdown.

The catch you'll hit once

The loader checks that the ID's manifest model_type is actually checkpoints. If you wrote diffusion_models in the manifest and try to load it here, you get a clear error saying the item has the wrong model_type - a deliberate guard, and a much nicer failure than the mysterious "file not found" you'd otherwise chase. So keep categories straight when you build the manifest.

Install and troubleshooting

Install comes from the pack: ComfyUI Manager search for "ComfyUI Model Batch Downloader", or:

cd ComfyUI/custom_nodes
git clone https://github.com/watarika/ComfyUI-Model-Batch-Downloader.git

then restart ComfyUI. No Python dependencies - just make sure aria2c is on the PATH that starts ComfyUI, since the download half depends on it.

The errors you'll actually meet: download_result has no item id 'x' means the ID doesn't match anything in the manifest (check your spelling, or that the download actually completed and wasn't skipped by a failure); a category-mismatch error means the model_type in the manifest doesn't line up with this loader. Both are the pack telling you exactly what to fix, which is more than most node suites manage.

Categorymodel/download/loaders

Inputs (2)

NameTypeDefaultDescription
download_resultDOWNLOAD_RESULT
idSTRING

Outputs (3)

NameTypeDescription
MODELMODEL
CLIPCLIP
VAEVAE