Shanghai Jumpspace Intelligent Technology Co., Ltd. recently announced a major upgrade to its image generation model, the Step-1X series, with the launch of the more powerful Step-1X-Medium version. This upgraded version has achieved significant improvements in several areas: based on the MMDit architecture, the generation speed has increased by over 30%; after targeted training, the new version has enhanced understanding capabilities and better consistency between images and text, resulting in more natural detail and texture in the generated images.
The Step-1X-Medium introduces a "Picture-to-Picture" feature, allowing users to simply upload an image and provide basic instructions to enhance details, apply style transfer, or make local modifications to the original image. Additionally, the new version has upgraded its ability to create "Chinese-style" content, better capturing the essence of Eastern facial features and presenting a more advanced and refined image texture. Furthermore, Step-1X-Medium supports adding English text in prompts, enabling the generated images to display English copy.
The upgraded Step-1X-Medium aims to be a powerful assistant for creators, deeply understanding the input ideas and providing more accurate and perfect output results. Currently, the new capabilities of Step-1X-Medium are available to users through API calls in the "Experience Center" of the Jumpspace open platform.
The new Step-1X-Medium has reached a new level in image generation quality, capable of producing more diverse scenes with stronger consistency between images and text. It can also deeply optimize Eastern character imagery, easily capturing the essence of Chinese style, generating consistently styled comic pages for fans of Chinese, Japanese, and American comics. For brand designers, Step-1X-Medium can generate advertising, product packaging, and marketing materials that align with brand tone, better showcasing the cultural core of the brand.
The "Base Image" feature launched with Step-1X-Medium allows creators to upload a base image, enabling the model to quickly understand the structure and style of the image and enhance details, transform styles, or refine specific areas based on the original creative concept. Additionally, Step-1X-Medium supports the SRef (Style Reference) generation function, providing style reference images from which the model extracts aesthetic styles and atmospheric features to incorporate into the composition of the generated images.
The advancements in AI technology allow Step-1X-Medium to add short English text in prompts, enhancing the visual artwork. This upgrade not only improves the quality and efficiency of image generation but also offers creators more creative space and possibilities.
Experience Link: https://platform.stepfun.com/
