Install Automatic1111 or Forge in Fedora using Anaconda

Written on the 3rd of May, 2024. Updated 7th of July, 2024.

Preamble

I wrote these instructions primarily for myself so that I can set things back up in case a hard-drive breaks... which totally didn't happen to me. So it's not exactly noob-friendly, but I figured there's no reason not to share it. I performed all of this in Fedora 40 but I don't see any reason you wouldn't be able to do this in any Linux-based distro, as it uses Anaconda to set up the environment which is platform independent. Most distro-specific stuff is related to your GPU drivers anyway which this guide assumes you've already got set up and running.

The reason I use conda is that it's completely separate from the system and is the only way I got this up and running inside of Fedora (there are no Fedora-specific instructions for any AUTOMATIC1111 workflows as of writing). Fedora does offer a python3.10 package along with its libraries but I haven't figured out why it doesn't include Python3.10 versions of any static packages required for AUTOMATIC1111 to run.

Prerequisites

Nvidia GPU. This guide does not cover AMD or Intel GPUs. Ensure that the Nvidia drivers and the Nvidia CUDA drivers are installed. This will vary greatly depending on your distribution.

Install the prerequisites:

1

sudo dnf install conda gperftools-libs libglvnd-glx

For other distros: https://github.com/AUTOMATIC1111/stable-diffusion-webui?tab=readme-ov-file#automatic-installation-on-linux (follow "Installing dependencies" step) and https://docs.conda.io/projects/conda/en/latest/user-guide/install/linux.html

If trying this on a distro other than Fedora, this is where you would install Python3.10 and then create the venv or run export PYTHON=/bin/python3.10 and skip ahead to "Run the Installer". If you are on Fedora continue reading.

Setting up conda

We will use conda to make a static, portable python environment. I'll be using a folder in ~/Applications/AUTOMATIC1111-Forge but feel free to use whatever path you want. Create a file named environment.yaml containing the following:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

name: forge
channels:
  - defaults
  - conda-forge
dependencies:
  - python=3.10
  - pip=23.0
  - packaging
  - gcc=12.1.0
  - gxx_linux-64

Open a terminal, cd to the directory of environment.yaml and then run:

1

conda env create -f environment.yaml

Wait a while for this to complete. Then:

1

conda activate forge

You'll notice that makes (forge) appear to the left of the input line in your terminal. You are now in your conda environment for forge.

Run the installer

If you haven't already, clone the repository to your drive. git clone https://github.com/lllyasviel/stable-diffusion-webui-forge. Replace this URL with the regular https://github.com/AUTOMATIC1111/stable-diffusion-webui if you're installing that instead.

If you already have your models, move them to stable-diffusion-webui/models/Stable-diffusion/ now to avoid redownloading the default one.

For AUTOMATIC1111: ./webui.sh --xformers
For Forge: ./webui.sh --cuda-stream
I you've ever launched Forge with -xformers enabled, you must use it or --disable-xformers for all future runs.

Everything should be automatically installed. This may take several hours to download and install everything. Once done, navigate to http://127.0.0.1:7860 in your web browser. Ctrl+C in your terminal to close the program, and conda deactivate to leave the conda environment. Or just close the terminal window.

Fix Controlnet

You'll notice if you try to use a controlnet model, it will fail. This is because insightface failed to be automatically installed. To fix this, close the webui and open a terminal in the stable-diffusion-webui folder, then run the following:

1
2

conda activate forge
source venv/bin/activate

You should have two environments active now, (venv) (forge). Run the following:

1

pip install insightface

It will complain about insightface requiring pydantic of at least >2.0.0, and you having 2.8.2. It's stupid. It installed successfully.

Set up the menu shortcut

To launch Forge from the start menu, create a file at /home/you/Applications/AUTOMATIC1111-Forge/AUTOMATIC1111-Forge.sh that contains the following (adjusting paths as necessary):

1
2
3
4
5
6
7
8

#! /bin/bash

# Retrieve the current BASH path in a string and tell conda where it is.
eval "$(conda shell.bash hook)"
# Activate the environment where a static Python is set up.
conda activate forge
# Start the web server.
cd /home/you/Applications/AUTOMATIC1111-Forge/stable-diffusion-webui-forge/ && ./webui.sh --xformers --update-check

Ensure that AUTOMATIC1111-Forge.sh is set as executable.

Create the desktop file to launch the program. I use KDE's Menu editor to create a shortcut with the following fields, though the only required field is the work path being set to where webui.sh is /home/you/Applications/AUTOMATIC1111-Forge/stable-diffusion-webui-forge/. Everything else left to its default/blank.

Name: AUTOMATIC1111 Forge
Description: SD Model WebUI
Comment: An extensible webUI for image generation and altering with Stable Diffusion. Forge is a fork with improved memory management and speed.
Program: /home/you/Applications/AUTOMATIC1111-Forge/AUTOMATIC1111-Forge.sh
Work path: /home/you/Applications/AUTOMATIC1111-Forge/stable-diffusion-webui-forge/
Run in terminal: Checked