On December 5, 2025, Jianfu Zhang, Zhiyuan CS alumni (Class of 2015); Assistant Professor, the School of Computer Science, SJTU, delivered a talk titled "AIGC Vision: Generation and Detection," engaging in in-depth discussions with the Zhiyuan students. With years of dedicated research in Artificial Intelligence Generated Content (AIGC) and trustworthy models, Zhang has published over 50 papers in top-tier conferences and journals such as ICCV, NeurIPS, ICLR, and CVPR, accumulating over a thousand citations.

Zhang Jianfu systematically outlined the technological evolution of visual generation models, tracing the journey from early breakthroughs like DALL·E 2 and Stable Diffusion to recent representative works such as Sora and Veo. He highlighted how AIGC has progressively pushed technical boundaries and reshaped the fundamental logic of content creation. These models, he noted, have not only achieved leaps in image generation quality but also demonstrated remarkable potential in video generation and multimodal understanding. They are profoundly transforming workflows across numerous fields, including artistic creation, media production, and education.

However, behind the rapid technological advancement, challenges related to authenticity and trustworthiness are becoming increasingly prominent. As the realism of AI-generated content continues to improve, traditional detection methods are gradually becoming obsolete, which makes preventing the spread of false information and rebuilding digital trust particularly urgent. In response, Zhang’s research team has focused its efforts on the direction of "AIGC Content Detection and Trust Verification."

To address the authenticity and trust challenges posed by AIGC, Zhang Jianfu’s research team is committed to developing technologies capable of accurately identifying AI-generated content. By deeply analyzing the inherent patterns and microscopic traces left during the content synthesis process of generation models, they have constructed a series of highly discriminative detection algorithms. These technologies can effectively distinguish between AI-generated content and authentic captured content, providing crucial technical support for scenarios such as content moderation, copyright protection, and forensic verification. During his presentation, Zhang emphasized that generation and detection are not opposing forces but rather complementary "technological wings" that evolve together. The future development of AIGC will inevitably be driven by their dynamic balance.

During the Q&A session, students raised questions regarding trends in visual and language models, model construction, and other related topics. Drawing from the current state of the field and his research experience, Zhang provided systematic responses and offered concrete suggestions on learning paths and practical directions. Following the discussion, many students stayed behind to engage in more in-depth conversations with Zhang on topics of personal interest.