< Explain other AI papers

TC-Light: Temporally Consistent Relighting for Dynamic Long Videos

Yang Liu, Chuanchen Luo, Zimo Tang, Yingyan Li, Yuran Yang, Yuanyong Ning, Lue Fan, Junran Peng, Zhaoxiang Zhang

2025-06-24

TC-Light: Temporally Consistent Relighting for Dynamic Long Videos

Summary

This paper talks about TC-Light, a new video relighting model designed to change the lighting in long, moving videos in a smooth and realistic way.

What's the problem?

The problem is that changing the lighting in videos often causes flickering or inconsistent results because the light effects can suddenly jump between frames, making the video look unnatural.

What's the solution?

The researchers made a two-stage model that first creates special features called appearance embeddings to represent how the video looks, then uses a unique optimization technique on video data to keep the lighting consistent over time while keeping the process fast and efficient.

Why it matters?

This matters because it helps filmmakers, game developers, and virtual reality creators make videos with changing lighting that looks natural and smooth, improving the overall quality and experience.

Abstract

TC-Light, a novel two-stage video relighting model, achieves high temporal coherence and low computational cost through appearance embedding and unique video tensor optimization.