Researcher · maker · listener · reader
I’m Yuqi Wang (王宇奇)—a visual intelligence researcher, an R&B and hip-hop listener, and a reader of philosophy.
One life, many frequencies: models, music, and questions.
Researching visual intelligence
Listening closely · Reading widely
01 / Selected publications
My work centers on one question: how can generative models follow complex human intent while preserving everything that should remain untouched?
Shot-wise keyframe anchors
The first controllable framework for multi-shot video editing, combining shot-aware planning, in-context keyframe editing, and edited-frame anchors. Unicorn improves the strongest baseline by 8.94% on MultiShotBench while preserving cross-shot identity.
Manuscript available after review
One model. One cognitive loop.
A unified self-rewarding framework that aligns instruction understanding with reward evaluation, reducing semantic drift in reinforcement-learning-based image editing.
Final paper coming soon
Generate what each client lacks.
Uses model prediction discrepancy to discover missing local data and guide diffusion-based generation, improving heterogeneous federated learning by up to 3.82%.
Read paper02 / Experience
From long-form video understanding to controllable generation, I enjoy carrying ideas all the way from a research question to a working system.
AIGC Algorithm Intern · Commercial Technology, Safety & Ecosystem
Leading research on multi-shot video editing and building MultiShotBench, an evaluation suite spanning instruction following, shot-level fidelity, visual quality, and motion.
AIGC Algorithm Intern · Digital Human & Creative Generation
Developed multi-element image composition and an end-to-end long-video auto-editing pipeline. The production system processed 1,000+ videos daily, improved editing efficiency by 8×, and lifted completion rate by about 12%.
03 / About
I’m a master’s student in Software Engineering at Beihang University, advised by Prof. Jianwei Niu in the LDMC Lab. My research spans diffusion and flow-matching models, multimodal language models, and reinforcement learning for visual generation.
Before graduate school, I completed my bachelor’s degree at Beihang, where I was named an Outstanding Graduate and received national first prizes in mathematical modeling and compiler design.
Outside the lab, I spend time with R&B and hip-hop—drawn to rhythm, texture, and voice—and with philosophy, returning to questions of mind, freedom, meaning, and how to live.
Precise changes with strong identity and structure preservation.
Connecting visual understanding with faithful generative decisions.
Better training signals for models that understand nuanced intent.
Beihang University · LDMC Lab
Beihang University · Outstanding Graduate
04 / Interests
Research is one way I make sense of the world. Music and philosophy are two others.
Groove, texture, sampling, and the way a voice can carry a lived world.
Questions of mind, freedom, meaning, and the relationship between thought and action.
应无所住，
而生其心
A line I return to: remain open, and let the mind arise without clinging.
05 / Personal blog
A future home for research notes, music, philosophy, and ideas worth keeping.
Personal notes · Coming soon
I’ll use this space to document experiments, research process, records I’m listening to, books I’m reading, and the occasional thought outside the lab.
Get in touch