Downloads and Paths

Downloads and Paths

Download backends

Backend	Use it when	Pause/resume
Python	You want the built-in downloader with no external executable.	No
aria2	You download large files and need segmented transfers or pause/resume.	Yes
Hugging Face Xet	Model Resolver detects Xet metadata for an eligible Hugging Face file.	No

Python is the default backend. Read Aria2 Download Backend before switching to aria2. Hugging Face Xet starts through the Python backend when the installed Hugging Face packages expose the required transport. Model Resolver falls back to the Python downloader when Xet cannot run.

Queue and download history

The queue panel separates workflow links from file transfers:

• Queued contains local matches waiting for Apply.
• Downloads > Active contains running, paused, cancelled, and failed transfers while the dialog tracks them.
• Downloads > History keeps completed download entries in the browser.

Use Download All With Sources for missing models that already have a trusted source. Use Search All Missing or a provider-specific batch action when sources are missing. A completed download can refresh its local match so you can apply the resolved path to the workflow.

Cancelling a Python download removes its partial file. Cancelling an aria2 transfer also removes the aria2 control file. Xet uses a temporary .xet-part file and removes it after cancellation.

Choose a target folder

Each download has a model category, root folder, and optional subfolder. The Download Path Mode setting controls how Model Resolver fills the subfolder:

• Suggest existing folders uses matching folders already visible to ComfyUI.
• Use path templates builds a subfolder from model metadata.
• Manual only leaves the field empty until you select or type a folder.

Per-download folder and subfolder fields override the global defaults. Default root settings let you select a preferred ComfyUI model root for supported categories.

Path templates

Templates accept these placeholders:

Placeholder	Value
{base_model}	Detected architecture, such as SDXL or Flux.
{author}	Provider creator or repository owner.
{first_tag}	The first useful tag, such as style or character.
{model_name}	Clean model name or file stem.
{version_name}	Provider version name when available.

Model Resolver sanitizes each path segment and keeps the result relative to the selected model root. Base Model Path Mappings can map a detected value to a folder path, such as Pony to SDXL/Pony.

Custom URLs

Use Custom URL when you have a direct file link. Choose the model category, target root, subfolder, and filename before starting the download. Model Resolver validates the URL and target path before it writes the file.