OmniGen Dataset Overview

1. Introduction

• Name: OmniGen
• Type: Unified image‑generation model
• Features: Supports multimodal prompts for image generation without extra plugins or preprocessing steps
• Goal: Provide a simple, flexible image‑generation paradigm

2. Core Capabilities

• Text‑to‑Image Generation
• Theme‑Driven Generation
• Identity‑Preserving Generation
• Image Editing
• Conditional Image Generation
• Reference‑Expression Generation (automatically detects objects in the input image)

3. Technical Highlights

• Methodology: See paper arXiv:2409.11340
• Advantages: Automatic extraction of input image features (objects, human pose, depth maps, etc.)
• Flexibility: New capabilities can be added via fine‑tuning

4. Resources

• Model Weights: Shitao/OmniGen‑v1
• Demo Platforms:
  • Hugging Face Demo
  • Replicate Demo
• Dataset: X2I Dataset

5. Usage

• Installation:

  git clone https://github.com/VectorSpaceLab/OmniGen.git
  cd OmniGen
  pip install -e .
  

• Quick‑Start Example: Provides examples for text‑to‑image and multimodal‑to‑image generation
• Diffusers Integration: Supports usage through the Diffusers library

6. Fine‑Tuning Support

• LoRA Fine‑Tuning
• Full‑Fine‑Tuning Options
• Training Script: train.py
• Guide: docs/fine‑tuning.md

7. License

• License: MIT License

8. Citation

@article{xiao2024omnigen,
  title={Omnigen: Unified image generation},
  author={Xiao, Shitao and Wang, Yueze and Zhou, Junjie and Yuan, Huaying and Xing, Xingrun and Yan, Ruiran and Wang, Shuting and Huang, Tiejun and Liu, Zheng},
  journal={arXiv preprint arXiv:2409.11340},
  year={2024}
}