In this tutorial, we introduce recent progress in low-level visual signal processing and quality assessment with generative AI and preference optimization. These approaches aim to leverage the latest generative foundation models, including diffusion models, large language models (LLMs), and multimodal LLMs (MLLMs), together with preference-driven optimization frameworks to address several pivotal challenges in low-level vision.

Real-world multimedia data typically suffer from low illumination, haze, blur, and adverse weather, severely degrading visual quality and downstream perception. Advancing low-level restoration is therefore essential for reliable interpretation in consumer imaging, surveillance, autonomous systems, and aerial platforms.

This special issue welcomes datasets and benchmarks, processing methods, and scalable low-complexity codec–model co-design and edge-efficient systems, all backed by transparent, reproducible evaluation. In summary, this special session seeks recent, significant advances in imaging, perception, and rea-soning for high-dimensional visual data, spanning the full pipeline from sensing and reconstruction to cross-modal understanding and efficient, trustworthy decision-making.

In this tutorial, we will introduce the recent progress of low-level image processing with Generative AI and Intelligent Agents, which aims to utilize the brand-new generative foundation models, e.g., diffusion models, large language models (LLMs) and multi-modal large language models (MLLMs) to tackle several pivotal challenges, e.g., perception quality, multi-modality information compensation, universality, automatic human-computer interaction (HCI) in the low-level image processing.