Skip to content
Xiang Wen

Xiang Wen

Tencent Hunyuan · Foundation Model Department

  • World model
  • Video generation
  • Diffusion models
  • Computer graphics
  • 3D vision
  • wenxiang at zju.edu.cn
  • TPAMI / ICMLJournal and conference
  • SIGGRAPH Asia / ACM MMGraphics and multimedia
  • GDC / GTCInvited talks
  • SkyReels-V12.7K stars

About

I am building EverWorld, real-time interactive spatial video models. You explore a scene, act on its objects and talk to its characters, and your input shapes what happens next. The world keeps generating while you are inside it, and what you changed is still there when you come back. It grows from one playable scene, to larger spaces you can compose and edit, to characters with their own memory and goals.

The problem sits across generative AI, diffusion models, computer graphics, and 3D vision. I have shipped work in all four, and the hard parts tend to fall in the seams between them. At Tencent Hunyuan I work on post-training for video generation: generation quality, a unified architecture handling text, image, multi-reference, and editing in one model, and the inference efficiency that real-time interaction ultimately depends on.

Before Tencent I built SkyReels AI, an AI short-drama platform, from scratch as its business lead and CTO, led game AI at ByteDance, and worked on game agents and content generation at NetEase Fuxi AI Lab. Over the same years I finished a PhD in artificial intelligence at Zhejiang University.

Read more →

News

  1. 2026.09Introduced EverWorld, real-time interactive spatial video models: explore a scene, act on its objects and talk to its characters, and your input shapes what happens next.Project
  2. 2026.03Received the SEVP Business Application Pioneer Award at Tencent for leading the algorithm work on AniMatrix.Award
  3. 2026.03HY-WU open-sourced at Tencent Hunyuan, a framework that generates instance-conditioned LoRA adapters on the fly for text-guided image editing.Open sourcearXiv
  4. 2025.12Completed my PhD in artificial intelligence at Zhejiang University.
  5. 2025.08VolGen accepted to TPAMI.TPAMI
  1. 2025.07Joined the Tencent Hunyuan foundation model department, working on post-training for video generation.
  2. 2025.02Open-sourced SkyReels-V1, a human-centric video foundation model, releasing both text-to-video and image-to-video weights along with the inference framework.Open source
  3. 2025.02Released SkyReels-A1, expressive portrait animation on video diffusion transformers, with code.arXivOpen source
  4. 2024.12Released Video Diffusion Transformers are In-Context Learners.arXiv
  5. 2024.09Talk at the Apsara Conference on AI workflows for producing drama online, from script generation to storyboarding.Apsara
  6. 2024.08Released SkyScript-100M, a script and shooting-script dataset for short drama.arXiv
  7. 2023.07Joined Kunlun Tech to build SkyReels AI, an AI short-drama platform, from scratch.
  8. 2023.03Talk at the GDC Machine Learning Summit on GPT-3 powered text to lifelike speech and animation for NPCs.GDC
  9. 2020.09Talk at the Beijing International Game Innovation Conference on AI-assisted game content production.BIGC
  10. 2019.12Talk at NVIDIA GTC China on video-guided intelligent choreography, shipped in Justice Online.GTC
Show all (15)Show less

Selected papers

All papers →
Show 11 moreShow less