On April 16, it was announced that MEITU Imaging Lab (MT Lab), a subsidiary of MEITU, has collaborated with the University of Bath to host the inaugural PortraitCraft Challenge at CVPR 2026 AI4VA, one of the top three conferences in the field of computer vision.

The competition features two tracks, with each winner eligible for a prize of up to $1,000. The first track emphasizes structured portrait composition understanding, evaluating whether models can accurately assess the compositional quality of portrait images. The second track focuses on whether models can genuinely interpret composition requirements, offering a more challenging and practical evaluation scenario for controllable portrait generation.

Additionally, PortraitCraft has released a large-scale, structured dataset tailored for portrait composition understanding and generation tasks. The dataset comprises approximately 50,000 carefully selected real portrait images, covering various typical composition scenarios such as single-person, multi-person, half-body, and full-body portraits. It includes multi-level annotations such as overall composition scores, 13 fine-grained composition attributes, explanatory texts, visual question-answering, and structured composition descriptions for generative tasks.

As MEITU's core research and development team, MEITU Imaging Lab has long focused on computer vision, image processing, and image generation, continuously advancing the application of cutting-edge AI technologies in imaging products. This challenge aims to leverage the lab's long-term expertise in imaging intelligence and aesthetic-related technologies to further promote research and development in portrait composition understanding and generation.