Arch Linux User Repository

Quick Start & Troubleshooting

How to run ComfyUI:

ComfyUI is packaged as a systemd service. The recommended way to run it is:

sudo systemctl start comfyui

Then visit http://localhost:8188 in your browser. To start it automatically on boot:

sudo systemctl enable --now comfyui

Do NOT run comfyui directly from your terminal as your regular user. The application files in /opt/comfyui are owned by the comfy system user. Running it as your own user will result in PermissionError: Permission denied on directories like /opt/comfyui/user, /opt/comfyui/temp, or /opt/comfyui/venv.

If you need to run it manually (e.g., for debugging), use:

sudo -u comfy comfyui

"Permission denied" on /var/lib/comfyui/input/3d:

This means the data directories were not created with proper ownership. Run:

sudo systemd-tmpfiles --create comfyui.conf

If the initial installation failed (e.g., PyTorch download interrupted):

The post-install setup may not have completed. Reinstall the package (through pacman -U or your preferred wrapper tool like yay or paru). This will re-run the setup, install dependencies, and apply correct ownership.

Custom install location:

To install ComfyUI to a different directory (e.g., a separate drive), build the package with:

env _prefix=/my/path makepkg -si

See also the README file in the repo

v0.21.1-7: Changed how build() handles the comfyui.install scriptlet when using a custom _prefix. Previously, sed replaced _PREFIX_ markers directly in $startdir, which meant the template had to be manually restored (git checkout comfyui.install) before rebuilding with a different prefix. build() now saves a no-clobber backup (.orig) and always generates from it, making repeated builds with different _prefix values work without manual intervention.

Some user instructions in comfyui.install file were also fixed (for instance where pacman -S was said to be used to reinstall, irrelevant for AUR).

Thanks for the reports RagingFlames and davispuh (0.21.1-5).

Fix: The default /etc/comfyui/extra_model_paths.yaml was fully commented out, causing yaml.safe_load() to return None and crash ComfyUI at startup with TypeError: 'NoneType' object is not iterable. An empty YAML mapping ({}) has been added so the file always parses to a valid value.

If you are already installed, the quickest fix without reinstalling is to add {} anywhere in your /etc/comfyui/extra_model_paths.yaml (e.g. as the first non-comment line). If you don't have any specific config that is.

I had same failure. To fix it needed to uncomment #comfyui: in /etc/comfyui/extra_model_paths.yaml

I just tried to install and run this. When I start the comfyui service it just terminates immediately. When I run as root as directed for testing, I get this.

$ sudo -u comfy comfyui
setup plugin alembic.autogenerate.schemas
setup plugin alembic.autogenerate.tables
setup plugin alembic.autogenerate.types
setup plugin alembic.autogenerate.constraints
setup plugin alembic.autogenerate.defaults
setup plugin alembic.autogenerate.comments
Traceback (most recent call last):
  File "/opt/comfyui/main.py", line 186, in <module>
    apply_custom_paths()
    ~~~~~~~~~~~~~~~~~~^^
  File "/opt/comfyui/main.py", line 111, in apply_custom_paths
    utils.extra_config.load_extra_path_config(config_path)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
  File "/opt/comfyui/utils/extra_config.py", line 10, in load_extra_path_config
    for c in config:
             ^^^^^^
TypeError: 'NoneType' object is not iterable

Thanks for the feedback pakar, here's what went in today (0.21.1-4):

GPU detection fix — The AMD GPU regex now also matches Display controller class devices, which covers APUs like the Strix Halo that don't show up as VGA or 3D.

Model paths config fix — The default /etc/comfyui/extra_model_paths.yaml had a base_path pointing to /var/lib/comfyui/models but was missing the per-type keys (checkpoints, loras, etc.), making it silently ineffective — ComfyUI was falling back to _prefix/models/ all along. After reflection, models are kept under the install prefix by design, since users control that location via the _prefix variable at build time (e.g. env _prefix=/large-drive/comfyui makepkg -si). The config file is now a commented-out template showing all supported model types, ready to uncomment and point to an alternate location.

Configurable input/output directories — A new /etc/comfyui/comfyui.conf file (sourced by the systemd service as an EnvironmentFile) lets you override the input and output directories without touching the service unit:

# /etc/comfyui/comfyui.conf
COMFYUI_OUTPUT_DIR=/my/drive/comfyui/output
COMFYUI_INPUT_DIR=/my/drive/comfyui/input

The file is listed in backup=() so pacman will preserve your changes on upgrades. Defaults remain /var/lib/comfyui/{output,input}.

Found a issue detecting the GPU on the system.

fix for comfyui.install:

 _detect_gpu() {
-    if lspci | grep -i "VGA\|3D" | grep -qi "AMD\|Radeon"; then
+    if lspci | grep -i "VGA\|3D\|Display controller" | grep -qi "AMD\|Radeon"; then

This is at least required for:

c3:00.0 Display controller: Advanced Micro Devices, Inc. [AMD/ATI] Strix Halo [Radeon Graphics / Radeon 8050S Graphics / Radeon 8060S Graphics] (rev d1)

Also, for comfyui to use the /var/lib/comfy folder you need a bit more configuration in /etc/comfyui/extra_model_paths.yaml

comfyui:
    base_path: /var/lib/comfyui/
    checkpoints: models/checkpoints/
    text_encoders: |
         models/text_encoders/
         models/clip/  # legacy location still supported
    clip_vision: models/clip_vision/
    configs: models/configs/
    controlnet: models/controlnet/
    diffusion_models: |
                 models/diffusion_models
                 models/unet
    embeddings: models/embeddings/
    loras: models/loras/
    upscale_models: models/upscale_models/
    vae: models/vae/
    audio_encoders: models/audio_encoders/
    model_patches: models/model_patches/
    is_default: true

Without this it will default to using /opt/comfyui/models/ instead /var/comfyui/models

Quick Start & Troubleshooting

How to run ComfyUI:

ComfyUI is packaged as a systemd service. The recommended way to run it is:

sudo systemctl start comfyui

Then visit http://localhost:8188 in your browser. To start it automatically on boot:

sudo systemctl enable --now comfyui

Do NOT run comfyui directly from your terminal as your regular user. The application files in /opt/comfyui are owned by the comfy system user. Running it as your own user will result in PermissionError: Permission denied on directories like /opt/comfyui/user, /opt/comfyui/temp, or /opt/comfyui/venv.

If you need to run it manually (e.g., for debugging), use:

sudo -u comfy comfyui

"Permission denied" on /var/lib/comfyui/input/3d:

This means the data directories were not created with proper ownership. Run:

sudo systemd-tmpfiles --create comfyui.conf

If the initial installation failed (e.g., PyTorch download interrupted):

The post-install setup may not have completed. Reinstall the package (through pacman -U or your preferred wrapper tool like yay or paru). This will re-run the setup, install dependencies, and apply correct ownership.

Custom install location:

To install ComfyUI to a different directory (e.g., a separate drive), build the package with:

env _prefix=/my/path makepkg -si

See also the README file in the repo

The permissions in the created directory structure seem to be incorrect. I get the same error as jawzper below; when I manually created a "/user" folder with admin access to Dolphin, it worked; it then errored on a missing "/temp" folder; created that. Started working at that point. I had to change the permissions on the models folder to be able to download models. Now for some reason despite my having changed the permissions on the "/output" folder, it can't create output files.

EDIT: It creates a folder structure under "/var/lib/comfyui" that also has only read-only permissions; changing all the folders there to read/write solved the output problem.