Shitong Shao

I am Shitong Shao (Sutton), a Ph.D. student in Data Science and Analytics at HKUST(GZ), advised by Prof. Zeke Xie. My research focuses on efficient generative modeling, with an emphasis on video diffusion, model distillation, efficient sampling, and production-ready generation systems.

My work spans the full path from method design to deployment. At First-Intelligence and Hedra, I contributed to real-world video-generation products, including HYVideo-1.5 and Character-C3. I am currently a Research Intern at ByteDance Seed, where I work on GAN design and distillation for image generation models.

中文主页 →

Research

Efficient generative modeling

I study video diffusion efficiency across step reduction, model compression, real-time inference, VAE acceleration, and sparse attention. Representative projects include MagicDistillation, FastLightGen, AMD, PISA, and LIVEditor.

Sampling and generation quality

I explore how initial-noise design and sampling-trajectory optimization can improve generation quality while reducing inference cost. Representative works include IV-Mixed Sampler, Golden Noise, and CoRe².

Research-to-production translation

I turn research ideas into deployable video-generation systems, with experience across training, systems optimization, serving, and product integration.

Selected publications

2026

1. ICML

   LIVEditor-14B: Lightning Unified Video Editing via In-Context Sparse Attention

   Shitong Shao, Zikai Zhou, Haopeng Li, and 4 more authors

   In ICML, 2026

   DOI HTML
2. IEEE TPAMI

   Improved and Accelerated Text-to-Image Generation With Collect, Reflect, and Refine

   Shitong Shao, Zikai Zhou, Dian Xie, and 5 more authors

   In IEEE TPAMI, 2026

   DOI HTML
3. CVPR

   FastLightGen: Fast and Light Video Generation with Fewer Steps and Parameters

   Shitong Shao, Yufei Gu, and Zeke Xie

   In CVPR, 2026

   DOI HTML
4. CVPR

   Accelerating Diffusion Model Training under Minimal Budgets: A Condensation-Based Perspective

   Rui Huang, Shitong Shao, Zikai Zhou, and 6 more authors

   In CVPR, 2026

   Rui Huang and Shitong Shao contributed equally.

   DOI HTML

2025

1. ICCV

   Golden Noise for Diffusion Models: A Learning Framework

   Zikai Zhou, Shitong Shao, Lichen Bai, and 4 more authors

   In ICCV, 2025

   DOI HTML
2. arXiv preprint

   MagicDistillation: Weak-to-Strong Video Distillation for Large-Scale Few-Step Synthesis

   Shitong Shao, Hongwei Yi, Hanzhong Guo, and 5 more authors

   In arXiv preprint, 2025

   DOI HTML
3. ICLR

   IV-Mixed Sampler: Leveraging Image Diffusion Models for Enhanced Video Synthesis

   Shitong Shao, Zikai Zhou, Lichen Bai, and 2 more authors

   In ICLR, 2025

   DOI HTML

2024

1. CVPR Highlight

   Generalized Large-Scale Data Condensation via Various Backbone and Statistical Matching

   Shitong Shao, Zeyuan Yin, Muxin Zhou, and 2 more authors

   In CVPR Highlight, 2024

   DOI HTML
2. NeurIPS

   Elucidating the Design Space of Dataset Condensation

   Shitong Shao, Zikai Zhou, and Zhiqiang Shen

   In NeurIPS, 2024

   DOI HTML

2023

1. IJCAI Oral

   Teaching What You Should Teach: A Data-Based Distillation Method

   Shitong Shao, Huanran Chen, Zhen Huang, and 3 more authors

   In IJCAI Oral, 2023

   DOI HTML

2022

1. ACCV Oral

   What Role Does Data Augmentation Play in Knowledge Distillation?

   Wei Li, Shitong Shao, Weiyan Liu, and 3 more authors

   In ACCV Oral, 2022

   Advisor as first author.

   DOI HTML

View all 56 publications →

Experience

Research Intern · ByteDance Seed
Apr. 22, 2026 – Present
Research on GAN distillation for efficient generative modeling.

Research Scientist Intern · First-Intelligence
Oct. 2025 – Apr. 1, 2026
Real-time generation and efficient deployment for HYVideo-1.5, including four-step video distillation and a 4.5×-accelerated VAE.

Research Scientist Intern · Hedra
Oct. 2024 – Jul. 2025
Built and productionized the distillation pipeline behind Character-C3 for few-step talking-video generation.

Research Intern · MBZUAI, Zhiqiang Shen’s Lab
Jul. 2023 – Mar. 2024
Worked on large-scale dataset condensation and optimization, resulting in G-VBSM (CVPR Highlight) and EDC (NeurIPS).

Research and Engineering Internships
Earlier roles at OPPO, Shanghai AI Laboratory, Beijing Institute of Technology, and Yiliu Technology covered dataset condensation, AI compilers, knowledge distillation, and model engineering.

Read the full CV →

Personal interests

Beyond research, I have a long-standing interest in traditional Chinese metaphysics and practice Liuyao divination, Qimen Dunjia, Meihua Yishu, and Four Pillars of Destiny (Bazi). My readings have been accurate roughly nine times out of ten—friends who are interested are welcome to reach out for a reading.