Transform Your Images with Step1X-Edit: Cutting-Edge Open Source Image Editing

Want to edit images like a pro without complex software or breaking the bank? Discover Step1X-Edit, a state-of-the-art, open source image editing model that rivals closed-source giants like GPT-4o and Gemini2 Flash. This innovative framework uses multimodal LLMs and diffusion image decoders to bring your vision to life, and it's all readily available.

Experience User-Friendly Image Editing with Real-World Results

Step1X-Edit isn’t just another research project; it’s designed for real-world use. Trained with a high-quality dataset and evaluated against the GEdit-Bench benchmark, it handles genuine user instructions with impressive accuracy. Download the inference code and model weights from ModelScope & HuggingFace and start editing images today.

Why Choose Step1X-Edit?

• Comparable Performance: Achieves results on par with leading proprietary models.
• Unified Image Editing: Excels at diverse editing tasks from simple adjustments to complex transformations.
• Open Source Advantage: Benefit from community-driven development and transparent algorithms.
• Real-World Focus: GEdit-Bench ensures evaluation reflects actual user editing needs.

Getting Started with Step1X-Edit

Ready to dive in? Here’s what you need to know to use this powerful image editing model.

System Requirements

• Hardware: H800 GPU (80GB memory recommended for optimal quality)
• Operating System: Linux
• Software: Python 3.10+

Installation Guide

1. Install Dependencies: pip install -r requirements.txt
2. Install Flash-Attention: Utilize the provided script to find the pre-built wheel suitable for your system: python scripts/get_flash_attn.py. Then, download and install the wheel. You might need to consult flash-attn documentation for detailed steps.

Run Your First Image Edit

With the dependencies installed, use bash scripts/run_examples.sh to run inference code on existing examples. This gives you an immediate overview of the tool’s abilities and allows instant understanding on the type of edits it can help with.

Evaluate and Adapt: Leverage the GEdit-Bench

GEdit-Bench is a novel benchmark that helps you to support a more authentic and comprehensive evaluation of image editing models. The evaluation process and related code are available in GEdit-Bench/EVAL.md.

Contributions and Acknowledgements

The Step1X-Edit team extends gratitude to the contributors of SD3, FLUX, Qwen, diffusers, and HuggingFace for their open research and exploration.

Important Disclaimer

Remember, the image editing results depend on user input, so the development team is not liable for any consequences arising from its use.