PixArt-Sigma is an open-source text-to-image generation model on GitHub from the PixArt-alpha team, succeeding the earlier PixArt-α model that is also listed in this directory. It is a Diffusion Transformer trained with a weak-to-strong strategy to produce images at up to 4K resolution, and the repository has collected about 1,900 stars.
PixArt-Sigma focuses on efficiency as much as quality. Its training approach starts from a weaker pretrained model and strengthens it with high-quality data and refined training techniques, achieving competitive image generation with far less training data and compute than comparable systems. The model generates images from text prompts at resolutions ranging from standard sizes up to 4K, with improved fine details, text rendering and prompt alignment over its predecessor. The official repository provides model weights, inference code and a project page showcasing sample generations, making it a solid foundation for research and downstream fine-tuning in the open image-generation ecosystem.
Alongside the model release, the team published detailed material on the training methodology, and community fine-tunes and integrations have since appeared across the open-source image generation ecosystem, including support in popular inference toolkits used by artists and developers.
PixArt-Sigma is free and open source under the Apache-2.0 license. The weights and code can be downloaded from GitHub and run on your own hardware, with no API fees, subscriptions or hosted-service charges from the project.
PixArt-Sigma is for image-generation researchers, model builders and developers who want an efficient open Diffusion Transformer that reaches 4K output, whether for academic study, fine-tuning domain-specific generators, or integrating open text-to-image models into products.