AI-assisted Long Video Creation

This workshop investigates how far current AI models and systems are from producing long-form, multi-shot videos that truly satisfy real users and creators. While recent generative models excel at short clips, long video creation introduces challenges in narrative structure, temporal consistency, multimodal alignment, and scalable human-in-the-loop editing.

A key motivation of this workshop is grounded in Bilibili’s large-scale creator ecosystem, where millions of creators produce narrative-driven, multi-minute videos spanning animation, education, entertainment, and storytelling. Such real-world production settings expose gaps between current academic benchmarks and practical creator needs, particularly in maintaining cross-shot consistency, narrative coherence, and efficient human–AI collaboration over long temporal horizons.

Leveraging production-grade data, creator workflows, and user feedback from the Bilibili platform, this workshop aims to bridge academic research with real-world impact. By bringing together researchers, industry engineers, and content creators, we seek to define technical roadmaps, creator-centered evaluation protocols, and reproducible benchmarks that measure not only visual quality, but also narrative satisfaction, usability, and audience engagement in AI-assisted long video creation.

We invite submissions that advance methods, datasets, systems, or evaluations for generating multi-shot, multi-minute videos that are coherent, controllable, and ethically responsible.

Topics of Interest

• Narrative and temporal consistency in long video generation
• Multimodal fusion and semantic alignment (video, audio, text, narration)
• Controllability, editing, and creator interfaces
• Knowledge-driven and genre-specific video creation
• Human-centric evaluation and benchmarking
• Human–AI collaboration and co-creation workflows

• Yudong Jiang (Bilibili) – Head of Video Foundation Model Technology
• Lisai Zhang (Bilibili) – Research Scientist
• Siqian Yang (Bilibili) – Research Scientist
• Baohan Xu (Bilibili) – Senior Research Scientist
• Yanzhang Liang (Bilibili) – Animation Production Executive
• Additional organizers from Bilibili and academia

The workshop will feature invited talks from leading researchers and practitioners in video generation, multimedia understanding, and AI-assisted creation.

• Zuxuan Wu (Fudan University) – Confirmed
  Associate Professor at Fudan University and member of the Fudan Vision and Learning Lab (FVL). His research focuses on large-scale video understanding, video generation, and efficient deep learning architectures.
• Qi Dai (Microsoft Research Asia) – Confirmed
  Principal Researcher at MSRA. His work spans video understanding and generation, including action recognition, representation learning, and large-scale multimedia intelligence.
• Xia Ning (Bilibili) – Tentative
  Senior animation director with extensive experience in film and television production. His works have received multiple national and international awards, and his interests include the integration of AI into animation pipelines.
• Sha Tong (Bilibili) – Tentative
  Leads AI products for creators at Bilibili, focusing on AI-powered video creation tools that enhance creator productivity and creative workflows.

Additional speakers may be announced.

AI-assisted long video creation has the potential to democratize storytelling and empower creators. At the same time, it raises concerns around misinformation, deepfakes, copyright, and labor impacts. This workshop explicitly encourages ethical analysis, mitigation strategies, and responsible deployment of generative video technologies.