Need help setting up ComfyUI on Fedora Linux 39

[michael-pfister]

I tried to set up ComfyUI on my local machine by following the instructions found in the README. On startup it seemed to be working fine, however when I tried to generate an image through the text prompt it spat out this confusing error message. This generative image stuff is a little new to me, sorry if I did something wrong.

got prompt
ERROR:root:!!! Exception during processing !!!
ERROR:root:Traceback (most recent call last):
  File "/home/michael/Documents/ComfyUI/execution.py", line 153, in recursive_execute
    output_data, output_ui = get_output_data(obj, input_data_all)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/Documents/ComfyUI/execution.py", line 83, in get_output_data
    return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/Documents/ComfyUI/execution.py", line 76, in map_node_over_list
    results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/Documents/ComfyUI/nodes.py", line 476, in load_checkpoint
    out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/Documents/ComfyUI/comfy/sd.py", line 439, in load_checkpoint_guess_config
    model_config = model_detection.model_config_from_unet(sd, "model.diffusion_model.", unet_dtype)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/Documents/ComfyUI/comfy/model_detection.py", line 157, in model_config_from_unet
    unet_config = detect_unet_config(state_dict, unet_key_prefix, dtype)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/Documents/ComfyUI/comfy/model_detection.py", line 50, in detect_unet_config
    model_channels = state_dict['{}input_blocks.0.0.weight'.format(key_prefix)].shape[0]
                     ~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
KeyError: 'model.diffusion_model.input_blocks.0.0.weight'

Prompt executed in 0.02 seconds
gc collect

These are the steps I used to get here:

# cloned the repo
git clone https://github.com/comfyanonymous/ComfyUI

# got a checkpoint file from https://civitai.com/models/42903/doll-likeness-by-edg and moved it to the appropriate directory
mv ~/Downloads/edgOlympiaDollLikeness.safetensors ~/Documents/ComfyUI/models/checkpoints/

# got a vae file from https://stable-diffusion-art.com/how-to-use-vae/#Download and moved it to the appropriate diectory
mv ~/Downloads/vae-ft-ema-560000-ema-pruned.ckpt ~/Documents/ComfyUI/models/vae

# installed python version 3.11 alongside with pip and switched alias since I had 3.12 installed
sudo dnf install python3.11.x86_64
alias python='/usr/bin/python3.11'
curl -sS https://bootstrap.pypa.io/get-pip.py | python3.11
alias pip='python -m pip'

# installed prerequisites and requirements
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm5.7
pip install -r requirements.txt

# executed main.py
python main.py
HSA_OVERRIDE_GFX_VERSION=10.3.0 python main.py # I also tried this since my GPU is a little older (2017)

These are my specifications:

             .',;::::;,'.                michael@fedora 
         .';:cccccccccccc:;,.            -------------- 
      .;cccccccccccccccccccccc;.         OS: Fedora Linux 39 (Workstation Edition) x86_64 
    .:cccccccccccccccccccccccccc:.       Kernel: 6.5.11-300.fc39.x86_64 
  .;ccccccccccccc;.:dddl:.;ccccccc;.     Uptime: 1 day, 19 hours, 19 mins 
 .:ccccccccccccc;OWMKOOXMWd;ccccccc:.    Packages: 2255 (rpm), 33 (flatpak) 
.:ccccccccccccc;KMMc;cc;xMMc:ccccccc:.   Shell: bash 5.2.21 
,cccccccccccccc;MMM.;cc;;WW::cccccccc,   Resolution: 3440x1440 
:cccccccccccccc;MMM.;cccccccccccccccc:   DE: GNOME 45.1 
:ccccccc;oxOOOo;MMM0OOk.;cccccccccccc:   WM: Mutter 
cccccc:0MMKxdd:;MMMkddc.;cccccccccccc;   WM Theme: Adwaita 
ccccc:XM0';cccc;MMM.;cccccccccccccccc'   Theme: Adwaita [GTK2/3] 
ccccc;MMo;ccccc;MMW.;ccccccccccccccc;    Icons: Adwaita [GTK2/3] 
ccccc;0MNc.ccc.xMMd:ccccccccccccccc;     Terminal: gnome-terminal 
cccccc;dNMWXXXWM0::cccccccccccccc:,      CPU: Intel i9-9900KF (16) @ 5.000GHz 
cccccccc;.:odl:.;cccccccccccccc:,.       GPU: AMD ATI Radeon RX Vega 56/64 
:cccccccccccccccccccccccccccc:'.         Memory: 8199MiB / 15893MiB 
.:cccccccccccccccccccccc:;,..
  '::cccccccccccccc::;,.                                                                                                      

During problem research I read somewhere that this is related to missing checkpoint files, so I checked permissions on the downloaded files edgOlympiaDollLikeness.safetensors and vae-ft-ema-560000-ema-pruned.ckpt but both have read permissions with my default user.

ls -l models/checkpoints/ models/vae
models/checkpoints/:
total 18544
-rw-r--r--. 1 michael michael 18986312 Nov 30 14:48 edgOlympiaDollLikeness.safetensors
-rw-r--r--. 1 michael michael        0 Nov 30 14:42 put_checkpoints_here

models/vae:
total 326852
-rw-r--r--. 1 michael michael         0 Nov 30 14:42 put_vae_here
-rw-r--r--. 1 michael michael 334695179 Nov 30 15:12 vae-ft-ema-560000-ema-pruned.ckpt

Thanks in advance!

[michael-pfister]

I noticed that I accidentally downloaded a LoRa file instead of a trained checkpoint which got fed into my "Load Checkpoint file". I figured it'd be OK since the readme mentioned to put files in models/checkpoints that have the .safetensors extension. After downloading an actual trained checkpoint 'v1-5-pruned-emaonly.ckpt' everything started working fine.