The human face, a dynamic canvas conveying nuanced emotions and unspoken messages, remains a captivating centerpiece in visual media. Recent generative AI methods have yet to fully master the detailed customization, indicating a need for further development in this area to achieve optimal precision and personalization in facial generative applications. The main objective of the research is to develop powerful AU-conditioned generation methods for efficient and precise generation and editing of facial expression images and videos with high fidelity and the capability to generate co-occurred AUs with different intensities, subtleness and fleetness, and spatial-temporal consistently among multi-modal input information, which is a very new and rarely explored topic. This can open new possibilities for face generation, amplifying the richness of human-human and human-AI interactions. The research is carried out in University of Oulu with the use of LUMI with collaborators from Italy and UK.

2025

2027