avatar

Junsong Chen

陈俊松

Ph.D. Student
HKU CDS
junsongc [at] icloud [dot] com


About Me

I am currently a research intern at NVIDIA Research supervised by Dr. Enze Xie and Prof. Song Han. I am also a Ph.D. student in MMLab at The University of Hong Kong under the supervision of Prof. Ping Luo. Previously, I have received my M.Phil. degree from IIAU Lab, advised by Prof. Huchuan Lu. My research area is the intersection of AIGC and machine learning, especially algorithm and system co-design for generative AI and its applications. I also build infrastructure for generative models, focusing on efficient inference systems and GPU kernel optimization.

News

Blogs

  1. Image
    Sol-H3: Speed-of-Light MiniMax-H3 Video Generation
    September 4, 2026
    MiniMax-H3 acceleration on NVIDIA Blackwell: H3 Super Acceleration generates 5s/10s 768p videos in just 6.85s/14.93s, and Sol-H3 goes further with native audio, reaching 3.00× realtime (a 5s, 1344×768 video in 1.669s). Sol-H3-Spark brings 5s, 768p video generation with audio to a single DGX Spark in 56s.
    Super Acceleration Sol-H3 Sol-H3-Spark
  2. Image
    Bet Small to Win Big: Efficient 2K Video Generation via Deeper-compression AutoEncoder, Linear Attention and Two-Stage Refiner
    February 11, 2026
    How to scale video generation from 720p to 2K without extra latency? This post introduces a Two-Stage paradigm: a high-compression base model for structure + a step-distilled refiner for details.
    Read Post MIT
  3. Image
    Infinite Context Length with Global but Constant Attention Memory
    November 20, 2025
    How does Linear Attention reduce complexity from O(N²) to O(N)? This post dives deep into the core idea of Linear Attention, derives its magical "state accumulation" property, and demonstrates how it empowers LLMs and long video generation.
    Read Post MIT

Selected Publications

Explore by topic
  1. arXiv
    📺 SANA-Video 2.0: Hybrid Linear Attention with Attention Residuals for Efficient Video Generation
    Junsong Chen, Jincheng Yu, Yitong Li, Shuchen Xue, Haozhe Liu, Jingyu Xin, Yuyang Zhao, Tian Ye, Zhangjie Wu, Zian Wang, Daquan Zhou, Ping Luo, Song Han, Enze Xie
    arXiv, 2026.
    Paper Project Page Code star

  2. arXiv
    ☀️ Sol-Attn: Accelerating Video Generation Inference via On-the-Fly Attention Sparsification
    Haopeng Li, Yitong Li, Junsong Chen, Tian Ye, Haozhe Liu, Jincheng Yu, Duomin Wang, Ruihua Zhang, Zeke Xie, Enze Xie, Song Han
    arXiv, 2026.
    Paper Project Page Code star

  3. arXiv
    ☀️ Sol Video Inference Engine: Agent-Native Full-Stack Acceleration Framework for Efficient Video Generation
    Yitong Li, Junsong Chen, Haopeng Li, Haozhe Liu, Jincheng Yu, Ligeng Zhu, Ping Luo, Song Han, Enze Xie
    arXiv, 2026.
    Paper Project Page Code star

  4. arXiv
    🌎 SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer
    Haoyi Zhu*, Haozhe Liu*, Yuyang Zhao*, Tian Ye*, Junsong Chen*, Jincheng Yu, Tong He, Song Han, Enze Xie
    arXiv, 2026. (* Equal contribution)
    Paper Project Page Models Code star

  5. arXiv
    🎬 SANA-Streaming: Real-time Streaming Video Editing with Hybrid Diffusion Transformer
    Yuyang Zhao*, Yicheng Pan*, Qiyuan He*, Jincheng Yu*, Junsong Chen*, Tian Ye, Haozhe Liu, Enze Xie, Song Han
    arXiv, 2026. (* Equal contribution)
    Paper Project Page Code star

  6. ICLR
    📺 SANA-Video: Efficient Video Generation with Block Linear Diffusion Transformer
    Junsong Chen*, Yuyang Zhao*, Jincheng Yu*, Ruihang Chu, Junyu Chen, Shuai Yang, Xianbang Wang, Yicheng Pan, Daquan Zhou, Huan Ling, Haozhe Liu, Hongwei Yi, Hao Zhang, Muyang Li, Yukang Chen, Han Cai, Sanja Fidler, Ping Luo, Song Han, Enze Xie
    ICLR, 2026. Oral
    Paper Project Page MIT YouTube Code star

  7. Image ICCV
    🏃SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation
    Junsong Chen*, Shuchen Xue*, Yuyang Zhao, Jincheng Yu, Sayak Paul, Junyu Chen, Han Cai, Song Han, Enze Xie
    ICCV, 2025. Highlight
    Paper Project Page MIT Code star

  8. Image ICML
    💪SANA-1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer
    Enze Xie*, Junsong Chen*, Yuyang Zhao†, Jincheng Yu†, Ligeng Zhu†, Yujun Lin, Zhekai Zhang, Muyang Li, Junyu Chen, Han Cai, Bingchen Liu, Daquan Zhou, Song Han
    ICML, 2025.
    Paper Project Page MIT Code star

  9. Image ICLR
    ⚡️SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer
    Enze Xie*, Junsong Chen*, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, Song Han
    ICLR, 2025. Oral
    Paper Project Page MIT Code star

  10. Image ECCV
    PixArt-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation
    Junsong Chen*, Chongjian Ge*, Enze Xie*†, Yue Wu*, Lewei Yao, Xiaozhe Ren, Zhongdao Wang, Ping Luo, Huchuan Lu, Zhenguo Li
    ECCV, 2024.
    Paper Project Page Youtube Code star

  11. Image ICML
    PixArt-δ: Fast and Controllable Image Generation with Latent Consistency Model
    Junsong Chen, Yue Wu, Simian Luo, Enze Xie, Sayak Paul, Ping Luo, Hang Zhao, Zhenguo Li
    ICML Workshop, 2024.
    Paper Project Page Demo Code star

  12. Image ICLR
    PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis
    Junsong Chen*, Jincheng Yu*, Chongjian Ge*, Lewei Yao*, Enze Xie, Yue Wu, Zhongdao Wang, James Kwok, Ping Luo, Huchuan Lu, Zhenguo Li
    ICLR, 2024. Spotlight / WAIC Youth Outstanding Paper Award, 2026
    Paper Project Page Demo Code star

  13. Image ICLR
    Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models
    Junyu Chen*, Han Cai*, Junsong Chen, Enze Xie, Shang Yang, Haotian Tang, Muyang Li, Yao Lu, Song Han
    ICLR, 2025.
    Paper Project Page Code star

  14. Image ICLR
    HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
    Haotian Tang*, Yecheng Wu*, Shang Yang, Enze Xie, Junsong Chen, Junyu Chen*, Zhuoyang Zhang, Han Cai, Yao Lu, Song Han
    ICLR, 2025.
    Paper Project Page Demo Code star

  15. Image ICCV
    MetaBEV: Solving Sensor Failures for BEV Detection and Map Segmentation
    Chongjian Ge*, Junsong Chen*, Enze Xie, Zhongdao Wang, Lanqing Hong, Huchuan Lu, Zhenguo Li, Ping Luo
    ICCV, 2023.
    Paper Project Page Youtube Code star

  16. Image CVPR
    ARKitTrack: A New Diverse Dataset for Tracking Using Mobile RGB-D Data
    Haojie Zhao*, Junsong Chen*, Lijun Wang, Huchuan Lu
    CVPR, 2023.
    Paper Project Page Youtube code star

Industry Projects

Our research SANA contributes to NVIDIA's flagship generative AI projects including DLSS and Cosmos. Besides, SANA-Sprint is in NVIDIA 3D AI Blueprint.

NVIDIA DLSS: Supreme Speed, Superior Visuals, Powered by AI
NVIDIA Cosmos: A World Foundation Model Platform for Physical AI
What are NVIDIA NIM Microservices and AI Blueprints for RTX AI PCs?

Talks

  1. Image
    SANA-Video: Efficient Video Generation with Block Linear Diffusion Transformer
    ICLR 2026 Oral Talk
    Oral presentation for SANA-Video at ICLR 2026.
    YouTube Project Page

  2. Image
    SANA-series
    Online Talk
    This talk introduces SANA-series, four different directions of acceleration for Efficient and high quality image & video generation.
    Qingke Bilibili TechBeat SANA-Series Project Page

Experiences

  1. Image
    NVIDIA, US remote
    Intern, working with Prof. Song Han and Dr. Enze Xie
    Project: SANA (Image/Video Generation, World Model)
    Project: Sol-Engine (Inference Engine)
    May 2024 - Present
  2. Image
    The University of Hong Kong, Hong Kong
    Research Assistant, working with Prof. Ping Luo and Dr. Enze Xie
    Project: PixArt (Image Generation)
    April 2023 - April 2024

Awards & Honors

Tencent Project Up Scholarship ( 15 students in China)
2026
KAUST AI Rising Star Award (25 speakers over the world)
2026
WAIC Youth Outstanding Paper Award for PixArt-α (10 papers selected)
2026
ECCV 2026 Outstanding Reviewer
2026
Outstanding Graduate of Liaoning Province
2022
Liaoning Provincial Table Tennis Singles Champion (Collegiate)
辽宁省大学生乒乓球单打冠军
2021
National Scholarship
2020

Services

Total homepage visits

Powered by Jekyll and Minimal Light theme.