Generative Vision Research

Tian (Owen) YE

PhD Student at HKUST,Guangzhou

Pushing the frontiers of Generative Vision. My work bridges state-of-the-art foundation models and scalable architectures to enable boundless human-AI co-creation.

Research AIGC
Core Models Diffusion Models
Long-term Bet Boundless Human-AI Co-Creation

About

Research that moves from generative foundations to boundless co-creation.

I work across generation, restoration, and perception, with an emphasis on methods that can travel from papers into open-source releases and product-facing systems.

I am a Ph.D. student working on Generative AI and Foundation Models. My work centers on building open-source systems that push the frontier of visual generation and restoration while being broadly reusable by the community.

I have led and contributed to influential open projects including LucidFlux, UltraFlux, PosterCraft, and PosterOmni, covering universal image restoration, native high-resolution generation, and creative design.

I also architected Meissonic, the first open-source non-autoregressive model to achieve SDXL-level performance, and co-architected MagicInfinite at Hedra, where it supported a product with strong commercial traction. My work has received international recognition, including selection as a KAUST Rising Star in AI (2025).

Research Highlights

Four threads that define the current research profile.

Open-source & Community Impact

Architected Meissonic, the first open-source non-autoregressive model to reach SDXL-level performance, with broad community adoption.

Visual Frontiers

Led UltraFlux (CVPR 2026) and LucidFlux (ICLR 2026), advancing the frontier of high-quality generation and restoration; LucidFlux surpasses strong commercial baselines such as Meitu SR.

Beyond Synthesis

Leveraged diffusion priors across restoration, perception, and creative design, including DTPM (CVPR 2024), AGLLDiff (AAAI 2025), GlassWizard (ICCV 2025), Posta (CVPR 2025), and PosterCraft (ICLR 2026).

Industry Translation

Co-developed MagicInfinite (Character-3) at Hedra for infinite talking-video generation, contributing to product traction and company growth ($15M ARR, $32M funding).

News

Recent updates from papers, releases, and talks.

2026.03

We release LucidNFT, the first consistency-driven RL paradigm for generative Real-SR.

2026.03

Our paper “Improved and Accelerated Text-to-Image Generation with Collect, Reflect, and Refine” is accepted by IEEE TPAMI.

2025.11

We release UltraFlux, a SOTA native 4K text-to-image generation model.

2025.09

We release LucidFlux-14B, a caption-free universal image restoration diffusion transformer.

2025.08

We release Flux.1-lite-8B-GRPO, an RL post-trained high-quality model based on Flux.1 Lite.

2025.06

Three papers are accepted by ICCV 2025.

2025.06

We release PosterCraft, a unified framework for high-quality aesthetic poster generation.

2025.03

We release MagicInfinite (Hedra Character-3), enabling fast infinite talking-video generation from words and voice.

2025.02

Three papers are accepted by CVPR 2025.

2025.02

We release Magic 1-For-1, a four-step image-to-video diffusion model, together with the technical report.

2024.11

Selected as an Outstanding Reviewer for BMVC 2024.

2024.11

We release Meissonic on Hugging Face, the first SDXL-level high-resolution non-autoregressive text-to-image model.

2024.09

Two papers are accepted by ECCV 2024.

Recent Projects

Representative projects across generation, restoration, and deployment.

Profile and Contact

Credentials, service, and current training.

Recognition

  • ICLR 2025 Notable Reviewer
  • KAUST AI Rising Star, 2025
  • Outstanding Reviewer, BMVC 2024
  • PG scholarship of HKUST(GZ), 2024

Education & Experience

  • PhD Student, HKUST(GZ), Aug 2024 to present
  • Research Scientist Intern, Hedra Inc., Nov 2024 to Aug 2025
  • Research Assistant, HKUST(GZ), Jun 2023 to Jul 2024
  • BEng, Jimei University, Sep 2019 to Jul 2023

Academic Services

Reviewer for ACCV, WACV, BMVC, AAAI, ICCV, CVPR, ECCV, ACM MM, NeurIPS, ICLR, and ICML. Workshop competition organizer for LOVEU@CVPR 2024.

Mentoring

Mentoring Song Fei, MPhil Student at HKUST(GZ).