lv-iberget/disco-diffusion-1

Disco Diffusion

A frankensteinian amalgamation of notebooks, models and techniques for the generation of AI Art and Animations.

Changes in this Fork

• Focus on running from Windows, Linux, Google Colab, and Docker via either CLI or Notebook.
• Move all functions possible out of the main disco.py module into dd.py so that disco.py can become readable.
• Instead changing parameters directly in disco.py, parameters can be controlled by environment variables, command-line arguments, or config (YAML) files.
• Allows YAML parameter files to be used as inputs. (See examples/configs for more)
• SQLite support to store parameters and images for querying against later.
• Iterate over parameters via modifiers and multipliers. (See examples/configs for more)

Windows First-time Setup (Anaconda)

Follow these steps for the first time that you are running Disco Diffusion from Windows.

Pre-requisites

• Anaconda installed
• Nvidia CUDA Toolkit Installed
• MS VS Community Installed with C++ checkmarked

1. From Anaconda Powershell Prompt:

   This command will allow you to use conda from a "regular" powershell session.

   conda init powershell
   exit
   

2. From your VS Code Powershell prompt:

   Note: These commands should be run from the working directory of this cloned repository.

   This command will pull all dependencies needed by Disco Diffusion into a conda environment called discodiffusion

   conda env create -f environment.yml
   conda activate discodiffusion
   

   • Note: If you have already activated a discodiffusion conda environment, you can refresh it if this repo has changed by typing:

     conda env update --prefix discodiffusion --file environment.yml --prune

3. Compile pytorch3d

   Note: These commands should be run from the working directory of this cloned repository.

   For reason I'm not 100% clear on, pytorch3d must be compiled in Windows. (Hence the requirement for C++ tool mentioned in Pre-requisties...)

   git clone https://github.com/facebookresearch/pytorch3d.git
   cd pytorch3d
   python setup.py install
   cd ..
   

4. Execute a test run:

   The following test run will run with all defaults (so "the lighthouse run" as it is coined.) Image output and current image progress (progress.png) will be stored in images_out.

   conda activate discodiffusion
   python disco.py
   

First-time Setup (pip)

• Windows PowerShell:

  python3 -m pip install --user virtualenv
  python -m venv .\.venv\discodiffusion
  .venv\discodiffusion\Scripts\Activate.ps1
  pip install -r requirements.txt

• Windows Command Prompt:

  python3 -m pip install --user virtualenv
  python -m venv .\.venv\discodiffusion
  .venv\discodiffusion\Scripts\Activate.cmd
  pip install -r requirements.txt

• Linux

  python3 -m pip install --user virtualenv
  python -m venv .venv/discodiffusion
  source .venv/discodiffusion/bin/activate
  pip install -r requirements.txt
  pip install git+https://github.com/facebookresearch/pytorch3d.git@stable

• Test Run

  python disco.py

  A batch run should begin. Kill with Control+C Example:

  Using device: cuda:0
  💻 Starting Run: TimeToDisco(0) at frame 0
  🌱 Randomly using seed: 1825817222
  Prepping models...
  🤖 Loading model 'ViT-B/32'...
  🤖 Loading model 'ViT-B/16'...
  🤖 Loading model 'RN50'...
  🤖 Loading secondary model...
  🤖 Loading LPIPS...
  🌱 Seed used: 1825817222
  Frame 0 📝 Prompt: ['A beautiful painting of a singular lighthouse, shining its light across a tumultuous sea of blood by greg rutkowski and thomas kinkade, Trending on artstation.', 'yellow color scheme']
  Batches:   0%|          | 0/50 [00:00<?, ?it/s]
  
  4%|████▌                                                                                                         | 10/240 [00:14<05:05,  1.33s/it]