👋 About Me
I am a Ph.D. student at HKUST(GZ), advised by Prof. Lei Zhu and Prof. Kan, Ge Lin. Previously, I worked as a Research Scientist Intern at Hedra Inc. My research is fueled by a core philosophy: to deliver work that is either immediately deployable by the user community or foundational enough to inspire the field. Guided by this goal, I strive to bridge the gap between academic innovation and industrial application.
🔬 Research Experience
- Community Utility: I released Meissonic, the first High-Res. Non-AR model achieving SDXL-level performance, which has garnered broad community adoption.
- Visual Frontiers: I lead UltraFlux (pioneering Data-Model Co-Design for Native 4K generation) and LucidFlux (a universal restoration model that outperforms commercial tools like Meitu SR), pushing the limits of visual quality and fidelity.
- Beyond Synthesis: I explore the versatile potential of generative models beyond standard synthesis. By harvesting Diffusion Priors, I have developed novel solutions for restoration (DTPM, CVPR 2024; AGLLDiff, AAAI 2025), perception (GlassWizard, ICCV 2025), and creative design (Posta, CVPR 2025, PosterCraft), validating the broad applicability of AIGC foundations.
- Industry Impact: At Hedra, I co-developed the MagicInfinite (Character-3) model. This breakthrough in infinite talking video generation directly propelled the company to $15M ARR and secured $32M funding.
📰 News
- [2025-11] We release UltraFlux: Data-Model Co-Design for High-quality Native 4K Text-to-Image Generation across Diverse Aspect Ratios. A SOTA Native 4K Generation Model!
- [2025-09] We release LucidFlux-14B:Caption-Free Universal Image Restoration via a Large-Scale Diffusion Transformer. A SOTA Universal Image Restoration DiT Model!
- [2025-08] Our Style LoRA series for the FLUX.1 Kontext model has surpassed 30K downloads and 100+ likes on Hugging Face！🎉🎉 Demo and LoRAs.
- [2025-08] MovieChat+ is accepted by IEEE TPAMI 2025.
- [2025-08] I am pleased to release Flux.1-lite-8B-GRPO. A RL Post-training high-quality model based on Flux.1 Lite.
- [2025-06] 3 Papers are accepted by ICCV 2025.
- [2025-06] We are pleased to announce the release of PosterCraft! A Unified Framework for High-Quality Aesthetic Poster Generation.
- [2025-03] We are pleased to announce the release of MagicInfinite (Character-3 Model of Hedra Inc.). Now you can fastly generate infinite talking videos with your words and voice!
- [2025-02] 3 Papers are accepted by CVPR 2025.
- [2025-02] We are pleased to announce the release Magic 1-For-1, a SOTA, 4-step image-to-video diffusion model, along with our technical report.
- [2025-01] I am honored to be selected as a speaker at KAUST Rising Stars in AI Symposium 2025!! Thank you KAUST for the opportunity!
- [2024-11] I am honored to be selected as a Outstanding Reviewer for BMVC 2024！
- [2024-11] We release Meissonic on HuggingFace🎉, Meissonic-1B is the first SDXL level, high-resolution non-AR T2I model!!
- [2024-09] 2 Papers are accepted by ECCV 2024.
Mentoring
- Song Fei, Mphil Student@HKUST(GZ)
🎖 Competitions & Awards
- ICLR 2025 Notable Reviewer
- KAUST AI Rising Star, 2025
- Outstanding Reviewer, BMVC 2024
- PG scholarship of HKUST(GZ), 2024
- 2022 CVPR NAS Competition Supernet Track: Third Place (Track 1)
- 2022 JMU Student Star Award (20/19000)
💬 Academic Services
- Conference Reviewer: ACCV 2022&2024, WACV 2023&2024, BMVC 2023&2024, AAAI 2022&2023&2024&2025, ICCV 2023, CVPR 2024, ECCV 2024, ACM MM 2024, NeurIPS 2024, ICLR 2025, ICML 2025.
- Workshop Competition Organizer: LOVEU@CVPR 2024
- Journal Reviewer: International Journal of Computer Vision IEEE Transactions on Image Processing IEEE Signal Processing Letter IEEE Journal of Oceanic Engineering
📖 Educations & Experience
- Aug'2024–Present: PhD Student, The Hong Kong University of Science and Technology, Guangzhou
- Nov'2024–Aug'2025: Research Scientist Intern, Hedra Inc.
- Jun'2023–Jul'2024: Research Assistant, The Hong Kong University of Science and Technology, Guangzhou
- Sep'2019–Jul'2023: B.Eng (Telecommunication Engineering), Jimei University, Xiamen