Skip to content

Aria2 Download Backend

Dariusz L edited this page Aug 3, 2026 · 3 revisions

ComfyUI Model Resolver can use aria2c as an optional download engine for model files. The default Python downloader still works without any extra setup, but aria2 is useful when you download large files and want better control over active transfers.

What aria2 adds

  • Pause and resume support for downloads started through Model Resolver.
  • Better handling of large files and interrupted transfers.
  • Parallel/segmented downloading when the remote server supports it.
  • A separate download process, so the UI can stay responsive while large files are being downloaded.

Setup with the built-in installer

  1. Open Model Resolver.
  2. Go to Options.
  3. Find Download Engine.
  4. Change the backend to aria2.
  5. Click Install aria2.
  6. Wait until the status message says that aria2 was installed and detected.
  7. Click Save if you changed any other settings.

The installer downloads the official aria2 release for your platform when a matching prebuilt binary is available. It saves the executable inside this custom node directory under:

tools/aria2/

After installation, Model Resolver fills aria2c path automatically and uses that executable for future aria2 downloads.

Manual setup

If automatic installation is not available for your system, install aria2 and make aria2c available in your system PATH. Leave aria2c path empty in Model Resolver.

Model Resolver accepts an explicit path only for an executable installed inside its managed tools/aria2/ directory. The Options tooltip and backend validation enforce this restriction.

Windows

Download aria2 from the official release page:

https://github.com/aria2/aria2/releases

Extract the archive, then add the directory containing aria2c.exe to your system PATH. Open a new terminal and check it with:

aria2c --version

Linux

Install aria2 with your package manager, for example:

sudo apt install aria2

or:

sudo pacman -S aria2

Leave aria2c path empty when aria2c is available in your system PATH:

aria2c --version

macOS

If you use Homebrew:

brew install aria2

Leave aria2c path empty after Homebrew adds aria2c to your PATH. Check it with:

which aria2c

Checking the configuration

Use Check aria2 in Options. A valid setup should show that aria2 is available. If the daemon is not running yet, that is normal. Model Resolver starts the aria2 RPC process automatically when the first aria2 download begins.

How downloads behave

When the aria2 backend is active:

  • New model downloads are sent to aria2.
  • Pause and resume buttons work for aria2 downloads.
  • Cancel removes the active aria2 transfer.
  • Existing Python-download behavior remains available by switching the backend back to Python.

Troubleshooting

aria2c executable was not found

Model Resolver could not find aria2c.

Try one of these:

  • Click Install aria2 again.
  • Run Install aria2 again, or place aria2c in your system PATH.
  • Add aria2 to your system PATH.
  • Switch back to the Python downloader.

Automatic install is not supported on my platform

The installer only uses official prebuilt aria2 binaries. If the official release does not provide a suitable binary for your OS/architecture, install aria2 manually, place aria2c in your system PATH, and leave aria2c path empty.

CivitAI downloads fail with authorization or access errors

Some CivitAI downloads require a valid token or a resolved download URL. Make sure your CivitAI token is configured in Model Resolver options. If the same link works in the browser but not in aria2, switch temporarily to the Python backend and try again.

Downloads do not pause immediately

Pause/resume commands are sent to the aria2 RPC process. Busy downloads or remote server delays can take a moment to report the new state. The UI updates after aria2 confirms the transfer state.

Removing aria2 installed by Model Resolver

Close ComfyUI, then remove the downloaded aria2 folder from:

tools/aria2/

After that, either install aria2 again from Options or switch the download backend back to Python.

Clone this wiki locally