OSS Alternative - Discover Top Open Source Alternatives to Popular Software

Nerogar/OneTrainer

A comprehensive, one-stop solution for training various Diffusion models, offering extensive features for fine-tuning, dataset preparation, and model management.

Core Features

Supports a wide range of Diffusion models (SD, SDXL, Flux, PixArt, etc.) and formats (diffusers, ckpt).

Offers diverse training methods including full fine-tuning, LoRA, and embeddings.

Automated dataset preparation with captioning (BLIP, BLIP2) and mask creation (ClipSeg, Rembg).

Optimized training techniques like Aspect Ratio Bucketing, Multi-Resolution Training, and Noise Scheduler Rescaling.

Integrated GUI and CLI for training, model conversion, and real-time sampling.

Quick Start

git clone https://github.com/Nerogar/OneTrainer.git && cd OneTrainer && ./install.sh

Detailed Introduction

OneTrainer is an all-encompassing open-source platform designed to streamline the training process for a wide array of Diffusion models, including Stable Diffusion, SDXL, Flux, and PixArt. It provides a unified environment for full fine-tuning, LoRA, and embeddings, supporting both diffusers and ckpt model formats. With integrated dataset tooling for automatic captioning and masking, alongside advanced training optimizations like Aspect Ratio Bucketing and Multi-Resolution Training, OneTrainer empowers users to efficiently prepare data and achieve high-quality model outputs. Its dual GUI and CLI modes offer flexibility for both beginners and advanced users, making complex diffusion model training accessible and manageable.