Excited to share our work “Identity Preserving 3D Head Stylization with Multi-view Score Distillation” is accepted to #ICCV2025 🚀

We tackle domain adaptation of 3D GANs through ID-preserving stylization, offering insights into diffusion-to-GAN distillation.

Preprint: arxiv.org/abs/2411.13536

Introducing RoPECraft🚀: training-free motion transfer for video DiTs. Flow-warped RoPE + flow-matching + phase regularization ⇒ prompt and motion faithful videos.

Co-first authored w A. Berke Gokmen, Yigit Ekin. Supervised by Aysegul Dundar.
arxiv.org/pdf/2505.13344
berkegokmen1.github.io/RoPECraft