The rapid advancement of generative AI (GenAI) has transformed the creation of images and videos, enabling unprecedented levels of visual realism, diversity, and controllability. Yet one pressing issue is in how to properly assess the quality of AI-generated visual content as such new types of contents poses significant challenges that extend well beyond the scope of traditional image and video quality assessment. Whereas conventional approaches have largely focused on perceptual distortions such as blur, noise, compression artifacts, and transmission degradation, assessing generated contents calls for a broader spectrum of evaluation framework one that have to consider semantic fidelity, temporal coherence, physical plausibility, aesthetic quality, and exact alignment with human expectations and prompts. This keynote will examine the transition from traditional perception-driven quality assessment to a new paradigm for AI-generated visual content, in which human perceptual experience needs to be integrated with principled understanding of semantics, motion, and the physical world. We shall highlight some recent efforts toward interpretable and comprehensive evaluation frameworks that assess not only visual fidelity, but also motion realism, multidimensional quality, relational semantic consistency, and physical plausibility. This talk will also discuss potential future directions for next-generation GenAI quality assessment, emphasizing the critical intersection of perception and principles as the foundation for designing trustworthy generative visual systems.

As multimedia experiences expand beyond conventional video streaming into AR/XR, automotive displays, immersive telepresence, and generative graphics, quality metrics must evolve to remain relevant. This talk argues that the next generation of quality assessment should be built on three pillars: psychophysics, optimisation, and graphics. First, psychophysical models and data provide the foundation for metrics that reflect how humans actually perceive distortions, rather than how signals differ in a purely mathematical sense. A perceptually grounded metric must account for the visual system's nonlinearities, attentional limits, and sensitivity to display characteristics such as resolution, luminance, contrast, and viewing geometry. Second, metrics must be evaluated not just on benchmark datasets but as objectives in real optimisation problems, where their loss landscapes reveal whether they are suitable for parameter tuning and system design. Third, computer graphics introduces a rapidly growing class of media and artifacts, from rendered imagery and neural representations to mixed-reality content, that differs substantially from traditional compression and transmission distortions. Metrics must therefore be robust to new artifact types and generalize beyond the datasets on which they were trained or validated. Together, these three pillars suggest a broader agenda for future media quality: metrics should be human-aware, optimisation-ready, and robust to the visual diversity of emerging graphics-driven experiences

There is no experience quite like interacting with a virtual human in an immersive environment. Even knowing no one is really there, you cannot help reacting as if they were, a pull most people first feel in immersive films or games. But can these virtual encounters create genuine impact in the real world beyond entertainment, and what makes them work? In this talk, I will present studies from our lab in which virtual encounters produce measurable real-world outcomes: training doctors in patient communication, reducing performance anxiety, and shifting attitudes on bias and climate change. Throughout, I'll reflect on how the quality and design of the immersive experience shape the effects we can achieve.