This repository provides a guide and scripts for running the latest Disco Diffusion AI art generation models (v5 Turbo with 3D animation) locally on Windows, bypassing the need for cloud platforms like Colab. It targets users who want to leverage advanced AI art generation on their own hardware, offering a more controlled and potentially cost-effective workflow.

How It Works

The project leverages the Windows Subsystem for Linux (WSL2) to create a Linux environment within Windows, enabling the use of Linux-only dependencies. It guides users through installing Anaconda within WSL2 for package management, setting up a specific PyTorch environment (v1.10 with CUDA 11.1) compatible with pytorch3d, and then running Disco Diffusion via Jupyter notebooks or an experimental command-line batch mode.

Quick Start & Requirements

• Install: Follow the detailed steps in the README, which involve installing WSL2, Anaconda within Ubuntu, creating a Conda environment, and installing dependencies.
• Requirements: Nvidia GPU with at least 8GB VRAM (12GB+ recommended), Windows 10 (version 21H2 or later) or Windows 11.
• Setup: Requires significant setup time due to WSL2 and Anaconda installation.
• Links: Official Microsoft WSL2 guide: https://learn.microsoft.com/en-us/windows/wsl/install

Highlighted Details

• Supports Disco Diffusion v5 Turbo with 3D animation capabilities.
• Includes an experimental batch mode for generating multiple videos with different prompts in a single run.
• Provides detailed troubleshooting advice for common CUDA errors, often related to VRAM limitations.
• Offers two methods for running the code: a Jupyter notebook interface and a command-line batch mode.

Maintenance & Community

The repository is maintained by MohamadZeina, who encourages users to open issues for troubleshooting and feature requests. There are no explicit mentions of other contributors, sponsorships, or a community forum (like Discord/Slack) in the README.

Licensing & Compatibility

The README does not explicitly state a license. The project's nature involves using Disco Diffusion, which itself has its own licensing. Compatibility for commercial use would depend on the underlying Disco Diffusion license and any other dependencies.

Limitations & Caveats

The batch mode is explicitly labeled as experimental and intended for video generation, not images. Users may encounter VRAM limitations, requiring resolution reduction or disabling certain models. The setup process is complex and requires familiarity with Linux command-line operations within WSL2.