/OneTrainer

OneTrainer is a one-stop solution for all your stable diffusion training needs.

Primary LanguagePythonGNU Affero General Public License v3.0AGPL-3.0

OneTrainer

OneTrainer is a one-stop solution for all your stable diffusion training needs.

OneTrainer Discord

Features

  • Supported models: Stable Diffusion 1.5, 2.0, 2.1, SDXL, Würstchen-v2, Stable Cascade, PixArt-Alpha and inpainting models
  • Model formats: diffusers and ckpt models
  • Training methods: Full fine-tuning, LoRA, embeddings
  • Masked Training: Let the training focus on just certain parts of the samples.
  • Automatic backups: Fully back up your training progress regularly during training. This includes all information to seamlessly continue training.
  • Image augmentation: Apply random transforms such as rotation, brightness, contrast or saturation to each image sample to quickly create a more diverse dataset.
  • Tensorboard: A simple tensorboard integration to track the training progress.
  • Multiple prompts per image: Train the model on multiple different prompts per image sample.
  • Noise Scheduler Rescaling: From the paper Common Diffusion Noise Schedules and Sample Steps are Flawed
  • EMA: Train you own EMA model. Optionally keep EMA weights in CPU memory to reduce VRAM usage.
  • Aspect Ratio Bucketing: Automatically train on multiple aspect ratios at a time. Just select the target resolutions, buckets are created automatically.
  • Multi Resolution Training: Train multiple resolutions at the same time.
  • Dataset Tooling: Automatically caption your dataset using BLIP, BLIP2 and WD-1.4, or create masks for masked training using ClipSeg or Rembg.
  • Model Tooling: Convert between different model formats from a simple UI.
  • Sampling UI: Sample the model during training without switching to a different application.
  • AlignProp: A Reinforcement Learning method for diffusion networks from the paper Aligning Text-to-Image Diffusion Models With Reward Backpropagation

Installation

Installing OneTrainer requires Python 3.10. You can download Python here https://www.python.org/downloads/windows/. Then follow these steps:

Automatic installation

  • Clone the repository git clone https://github.com/Nerogar/OneTrainer.git
  • Run:
    • Windows: install.bat
    • Unix based systems: install.sh

Manual installation

  • Clone the repository git clone https://github.com/Nerogar/OneTrainer.git
  • Navigate into the cloned directory cd OneTrainer
  • Set up a virtual environment python -m venv venv
  • Activate the new venv:
    • Windows: venv\scripts\activate
    • Unix based systems: source venv/bin/activate
  • Install the requirements pip install -r requirements.txt

In some linux distribution, you might need to install libGL, for instance on ubuntu you will need to run:

sudo apt-get update
sudo apt-get install libgl1

Updating

Automatic update

  • Run update.bat or update.sh

Manual update

  • Pull changes git pull
  • Activate the venv venv\scripts\activate
  • Re-Install all requirements pip install -r requirements.txt --force-reinstall

Usage

To start the UI, run start-ui.bat. You can find a quick start guide here., and a more detailed overview of different topics here.

If you need more control, OneTrainer supports two modes of operation. Command line only, and a UI. All commands need to be run inside the active venv created during installation.

All functionality is split into different scrips located in the scripts directory. This currently includes:

  • train.py The central training script
  • train_ui.py A UI for training
  • caption_ui.py A UI for manual or automatic captioning and mask creation for masked training
  • convert_model_ui.py A UI for model conversions
  • convert_model.py A utility to convert between different model formats
  • sample.py A utility to sample any model
  • create_train_files.py A utility to create files needed when training only from the CLI
  • generate_captions.py A utility to automatically create captions for your dataset
  • generate_masks.py A utility to automatically create masks for your dataset
  • calculate_loss.py A utility to calculate the training loss of every image in your dataset

To learn more about the different parameters, execute <scipt-name> -h. For example python scripts\train.py -h

Contributing

Contributions are always welcome in any form. You can open issues, participate in discussions, or even open pull requests for new or improved functionality. You can find more information here

Before you start looking at the code, I recommend reading about the project structure here. For in depth discussions, you should consider joining the Discord server.

Related Projects

  • MGDS: A custom dataset implementation for Pytorch that is built around the idea of a node based graph.
  • StableTuner: Another training application for Stable Diffusion. OneTrainer takes a lot of inspiration from StableTuner and wouldn't exist without it.
  • Visions of Chaos: A collection of machine learning tools that also includes OneTrainer.