In July of this year, Step Celestial officially released the Step-1X series of image generation models. Recently, the Step-1X-Medium model in this series has undergone a significant upgrade, bringing substantial performance improvements and more diverse functionalities to the field of image creation.

This upgrade is based on the MMDit architecture, which has increased the generation speed of the Step-1X-Medium by over 30%. After targeted training, the model demonstrates a stronger ability to understand and respond to creators' needs, with a notable enhancement in the consistency between text and images. The generated images now feature more natural and detailed textures.

To meet a wider range of creative needs, the new version introduces a "base image enhancement" feature. Users can upload a base image and provide simple instructions, and the model will enhance details, transfer styles, or make local modifications based on the original image. This feature greatly enriches the possibilities for image creation, making image editing and modification more convenient and efficient.

Additionally, the Step-1X-Medium has seen significant improvements in creating content with a traditional Chinese style. The model can more accurately capture the essence of Eastern characters, producing more sophisticated and refined visual qualities. This is a major boon for fans of Chinese, Japanese, and American comics, as well as brand designers.

It's worth noting that the new version also supports adding English text in the instructions, allowing the generated images to include English elements, further broadening the application scenarios for creation.

The upgraded features of the Step-1X-Medium are now available on the Step Celestial open platform, and users can experience them through API calls. This upgrade not only enhances the speed and quality of image generation but also provides creators with more diverse tools and options, contributing to the further development of the image creation field.