This project provides a standalone, GUI-driven training environment for the Anima diffusion model, specifically supporting LoRA training. It targets users who need a streamlined and decoupled setup for fine-tuning Anima, offering a user-friendly interface built upon the robust sd-scripts implementation. The primary benefit is simplifying the complex process of LoRA training for Anima, making it more accessible on both Windows and Linux systems.

How It Works

The trainer is built upon the sd-scripts implementation, leveraging its core functionalities for diffusion model training. It focuses exclusively on LoRA (Low-Rank Adaptation) training for the Anima model, allowing users to efficiently fine-tune the model without retraining all parameters. The system features a web-based GUI, simplifying configuration and launch processes.

Quick Start & Requirements

• Primary install/run command: Clone the repository, then execute .\setup_env.bat (Windows) or ./setup_env.sh (Linux) to create a virtual environment and install dependencies. Launch the UI with .\training-ui\start_training_ui_anima.bat (Windows) or ./training-ui/start_linux.sh (Linux). The UI is accessible at http://localhost:3000.
• Non-default prerequisites: Python 3.10+ (3.12 recommended), Node.js (for Web UI), and a compatible CUDA installation (12.7+ recommended).
• Dependencies: The setup script installs Python dependencies, including PyTorch. Users may need to manually install specific PyTorch versions, e.g., pip install torch==2.7.0 torchvision==0.22.0 --index-url https://download.pytorch.org/whl/cu128.
• First-time setup: Requires configuring paths to the Anima DiT model, VAE, and CLIP text encoder within the Global Settings of the UI.
• Links: No direct links to official quick-start guides or demos are provided in the README.

Highlighted Details

• Linux support was introduced in v2.0.0, alongside multi-GPU inference capabilities.
• Version 1.1.0 focused on improving caching and I/O performance.
• The system has been tested on Windows 11 with high-end hardware configurations, including multiple GPUs (RTX 5080 + RTX 3090), 96GB DDR5 RAM, Python 3.12.1, CUDA 13.1, and PyTorch 2.10.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord or Slack), sponsorships, or a public roadmap are mentioned in the provided README.

Licensing & Compatibility

The README does not specify a software license. Therefore, compatibility for commercial use or closed-source linking cannot be determined from the provided text.

Limitations & Caveats

Some features and settings available in the original sd-scripts may not be fully implemented or functional in this standalone trainer. Users might encounter issues with distributed training on Windows, with a suggested workaround involving setting the GLOO_SOCKET_IFNAME environment variable. Specific PyTorch and CUDA version compatibility can be sensitive; the project notes it works best with torch<=2.3 and cuda<=12.4 without applying specific fixes.