HINT-SD: Targeted Hindsight Self-Distillation for Long-Horizon Agents
Woongyeng Yeo^*, Yumin Choi^*, Taekyung Ki, Sung Ju Hwang^†
Empirical Methods in Natural Language Processing (EMNLP Findings), 2026
arXiv / Code / Hugging Face

Targeted self-distillation framework that selectively distills failure-relevant actions from full trajectories for more effective long-horizon agent training.

Self-Refining Video Sampling
Sangwon Jang^*, Taekyung Ki^*, Jaehyeong Jo^*, Saining Xie, Jaehong Yoon^†, Sung Ju Hwang^†
International Conference on Machine Learning (ICML), 2026
Project Page / Paper / arXiv / Code / Hugging Face

Improving physical plausibility in video generation without any additional training or external verifiers.

Avatar Forcing: Real-Time Interactive Head Avatar Generation for Natural Conversation
Taekyung Ki^*, Sangwon Jang^*, Jaehyeong Jo, Jaehong Yoon, Sung Ju Hwang^†
Conference on Computer Vision and Pattern Recognition (CVPR), 2026
Project Page / arXiv / Paper / Video / Code / Hugging Face

User-interactive head generation model based on motion latent diffusion forcing and a direct preference optimization method for interactive motion.

Frame Guidance: Training-Free Guidance for Frame-Level Control in Video Diffusion Models
Sangwon Jang^*, Taekyung Ki^*, Jaehyeong Jo, Jaehong Yoon, Soo Ye Kim, Zhe Lin, Sung Ju Hwang^†
International Conference on Learning Representations (ICLR), 2026
Project Page / arXiv / Paper / Code / Hugging Face

FLOAT: Generative Motion Latent Flow Matching for Audio-driven Talking Portrait
Taekyung Ki, Dongchan Min, Gyeongsu Chae
International Conference on Computer Vision (ICCV), 2025
Audio-Visual Generation & Learning Workshop at ICCV (AVGen@ICCV), 2025
Project Page / arXiv / Paper / Code / Hugging Face

Motion latent flow matching for real-time audio-driven talking portrait generation and editing using its learned orthonormal motion basis.

Learning to Generate Conditional Tri-plane for 3D-aware Expression Controllable Portrait Animation
Taekyung Ki, Dongchan Min, Gyeongsu Chae
European Conference on Computer Vision (ECCV), 2024
Project Page / arXiv / Paper / Supp

We propose a contrastive pre-training framework for appearance-free facial expression hidden in 3DMM and 3D-aware, expression-controllable portrait animation model.

StyleLipSync: Style-based Personalized Lip-sync Video Generation
Taekyung Ki^*, Dongchan Min^*
International Conference on Computer Vision (ICCV), 2023
Project Page / arXiv / Paper / Code / Supp

StyleLipSync can generate person-agnostic audio-lip synchronized videos by leveraging the strong facial prior of style-based generator.

Deep Scattering Network with Max-pooling
Taekyung Ki, Youngmi Hur^†
IEEE Data Compression Conference (DCC), 2021
Paper / Code

We mathematically prove that the pooling operator is a crucial component for translation-invariant feature extraction in Scattering Network.