CFG-Zero-star is an open source project focusing on improving Classifier Free Guidance (CFG) techniques in stream matching models, developed by Weichen Fan and the S-Lab team at Nanyang Technological University. The tool significantly improves the quality of image and video generation by optimizing the bootstrapping strategy and implementing a zero-initialization approach.
Key technical features of the project include:
- Improved traditional CFG techniques to enhance the match between generated content and input text
- A zero-initialization optimization method is introduced to improve the sample quality of the flow matching model
- Supports many popular generation models such as Stable Diffusion 3, SD3.5 and Wan-2.1.
Open-sourced under the Apache-2.0 license, the project is suitable for academic research as well as supporting commercial applications, and provides important technical support for the field of AI content generation.
This answer comes from the articleCFG-Zero-star: An Open Source Tool for Improving Image and Video Generation QualityThe
