Tencent’s Hunyuan team released a preview of Hy Image 3.5, its latest image-generation model for everyday and professional work.
The team's Tuesday note said the newest model is a continuation of the Hy Image line with improvements in text rendering and layout, photorealism and stylistic range, and editing consistency.
In an internal blind test with more than 100 professional designers, Tencent said the new model's overall win rate against the last version rose by more than 30%. External labs have not published matching scores.
Hy Image 3.5 preview supports text-to-image and image-to-image generation. For every request, users can upload up to five reference images and continue editing across multiple requests using conversation history. The preview model supports various multiple ratios at resolutions up to 2K.
Tencent said the model has been optimized for tasks that require accurate text, layered information, consistent subjects across edits, and natural lighting and materials.
Tencent's AI assistant and chatbot app Yuanbao began a limited rollout about a month before the public preview, according to the blog post. The model is now available on other Tencent AI applications, such as WorkRally, OnSolo, and Miora.
Hy Image 3.5 preview came out on the same day as Alibaba Group's new AI chip Zhenwu V900, which is expected to deliver three times the performance of its predecessor. These developments highlight China's push to build domestic alternatives to advanced AI technology in the U.S., especially after the U.S. blocked China's access to some high-end AI chips and chipmaking equipment through export controls.