Yeji Song

Research Scientist

ldynx.github.io/

ldynx

ldynx@snu.ac.kr

My primary focus is on video and image generation, editing and diffusion models. My recent works include:
• Expanding video editing capabilities by proposing a motion word, enabling subject replacement across diverse body structures.
• Identifying and resolving a conflict among embeddings in zero-shot text-to-image (T2I) customization using orthogonal embedding.
My research also includes a broader areas of computer vision, including video rendering, segmentation, and 3D object detection.

Education

Seoul National University
Ph.D. in Computer Science, advised by Prof. Nojun Kwak.
Supported by LG Electronics Scholarship 2023-2025
Mar 2020 - Feb 2026 (Expected)
Seoul, Korea
Seoul National University
B.S. in Statistics and Psychology (double-major).
Mar 2013 - Feb 2020
Seoul, Korea

Employment

Meta | Research Scientist Intern
• Developing methods to improve efficiency of image editing models by selecting semantically meaningful tokens.
• Establishing an end-to-end approach using lightweight MLPs based on Flux models.
Aug 2025 - Jan 2026
Seattle, WA
NAVER Webtoon | Research Scientist Intern
• Research on video generation model using the motion token concept. Published ECCV'24.
Jan 2023 - Jul 2023
Seongnam, Korea

Publications

[G] Image & Video Generation [S] Safe Generative Models [E] Model Efficiency [D] Object Detection
[G] Point-to-Point: Sparse Motion Guidance for Controllable Video Editing
Yeji Song, Jaehyun Lee, Mijin Koo, Nojun Kwak
ArXiv, 2025
[S] Targeted Data Protection for Diffusion Model by Matching Training Trajectory
Hojun Lee, Mijin Koo, Yeji Song, Nojun Kwak
AAAI, 2026
[G] ReFlex: Text-Guided Editing of Real Images in Rectified Flow via Mid-Step Feature Extraction and Attention Adaptation
Jimyeong Kim, Jungwon Park, Yeji Song, Nojun Kwak, Wonjong Rhee
ICCV, 2025 Highlight
[G] Harmonizing Visual and Textual Embeddings for Zero-Shot Text-to-Image Customization
Yeji Song, Jimyeong Kim, Wonhark Park, Wonsik Shin, Wonjong Rhee, Nojun Kwak
AAAI, 2025
[G] SAVE: Protagonist Diversification with Structure Agnostic Video Editing
Yeji Song, Wonsik Shin, Junsoo Lee, Jeesoo Kim, Nojun Kwak
ECCV, 2024
[S] Exploring Causal Mechanisms for Machine Text Detection Methods
Kiyoon Yoo, Wonhyuk Ahn, Yeji Song, Nojun Kwak
NAACL Workshop, 2024
[E] Finding Efficient Pruned Network via Refined Gradients for Pruned Weights
Jangho Kim, Jayeon Yoo, Yeji Song, Kiyoon Yoo, Nojun Kwak
ACM MM, 2023
[G] Towards Efficient Neural Scene Graphs by Learning Consistency Fields
Yeji Song, Chaerin Kong, Seoyoung Lee, Nojun Kwak, Joonseok Lee
BMVC, 2022
[D] Md3d: Mixture-density-based 3d object detection in point clouds
Jaeseok Choi, Yeji Song, Yerim Kim, Jaeyoung Yoo, Nojun Kwak
IEEE Access, 2022
[D] Part-aware data augmentation for 3d object detection in point cloud
Jaeseok Choi, Yeji Song, Nojun Kwak
IROS, 2021

Projects

Korean traditional art style transfer | Ministry of Culture, Sports and Tourism (MCST)
• Developed diffusion model that generates traditional Korean art styles.
• Built a benchmark to effectively compare baselines for Korean art style transfer.
Apr 2024 - Dec 2024
Video outpainting for Media Art | Easywith Co., Ltd.
• Addressed physical limitations (e.g., exhibition space mismatches) in the Media Art Exhibition.
• Developed a video outpainting model by extending a video scene into a panoramic image.
Nov 2023 - Feb 2024
Semantic segmentation for Webtoon images | Naver Webtoon Corp.
• Established a webtoon segmentation model as part of a webtoon translation project.
• Built a demo interface for segmenting and rearranging text balloons and typesets.
Jan 2023 - Mar 2023
Development of object detector for Fashion AI Chatbot | Ministry of Information and Communication Technologies (ICT)
• Participated in the development of Vision-Language Models (VLMs) for a Fashion AI Chatbot.
• Established an embedder for detecting clothing items and extracting critical features.
Jan 2022 - Dec 2022
Unmanned outdoor security robot | Samsung Electronics Co., Ltd.
• Enhanced the robot's detection capabilities using a limited amount of a real-world data.
• Improved mean average precision (mAP) from 39.8% to 91%.
Jun 2021 - Dec 2021

Reviewer

• AAAI 2026
• ICCV 2025
• CVPR 2025
• ACM MM 2023 - 2025

Teaching

• TA, Introduction to Diffusion-based Generative Models for The AI Lab (Feb 2024)
• TA, Linear Algebra for Hyundai Heavy Industries. (Mar 2022 - Jun 2024)
• TA, Mathematics for Intelligent Systems (Mar 2023 - Jun 2023)