Research

My research focuses on advancing generative AI by leveraging efficient one-step generation for real-world applications such as video enhancement and image inpainting. I am particularly interested in diffusion-based methods that achieve high-quality results with minimal computational cost, making generative models more practical and accessible. In addition to efficiency, I am also working on adversarial defenses to ensure the responsible and safe deployment of generative video models.

I aim to expand my research toward multimodal and video generation, with a focus on improving fine-grained temporal consistency and efficiency, while incorporating physically aware conditioning into these models.

Selected Preprints

* denotes equal contribution

A memory‑efficient adversarial attack against diverse image‑to‑video diffusion models, enabled by robust dual‑space perturbation optimization.

A highly efficient one-step inversion diffusion network for high-quality few-step image inpainting.

A plug-and-play, backbone-agnostic video enhancer with low compute: each frame requires only inversion plus one generator pass, making it easy to deploy on outputs from diverse T2V systems.

Selected Publications

* denotes equal contribution

iSM resolves five major shortcut-model flaws with dynamic guidance, wavelet loss, sOT, and Twin EMA, yielding markedly better image generation.

An improved SwiftBrush version that makes the one-step diffusion student beats its multi-step teacher.

A high-quality dataset centered on extreme pose faces, supporting face synthesis, reenactment, recognition benchmarking, and more.