Light4D

Training-Free Extreme Viewpoint 4D Video Relighting

School of Computer Science, Peking University
*Equal contribution. Project lead. Corresponding author.

30° Comparison

Wan2.1
EX-4D
EX-4D → LAV
LAV → EX-4D
IC-Light
Light-X
Light4D
Wan2.1
EX-4D
EX-4D → LAV
LAV → EX-4D
IC-Light
Light-X
Light4D
Wan2.1
EX-4D
EX-4D → LAV
LAV → EX-4D
IC-Light
Light-X
Light4D

90° Comparison

Wan2.1
EX-4D
EX-4D → LAV
LAV → EX-4D
IC-Light
Light-X
Light4D

180° Comparison

Wan2.1
EX-4D
EX-4D → LAV
LAV → EX-4D
IC-Light
Light-X
Light4D
Wan2.1
EX-4D
EX-4D → LAV
LAV → EX-4D
IC-Light
Light-X
Light4D

30° Visualization

Wan2.1 EX-4D Light4D

180° Visualization

Wan2.1 EX-4D Light4D

Different Angles

Wan2.1 30° 90° 180°

Method

Overview of the Light4D framework. Our training-free approach employs a time-aware paradigm in a latent flow-matching process. Using a multi-phase adaptive schedule λ(t), we prioritize 3D geometric completion via the EX-4D backbone before injecting illumination cues through IC-Light.

Design of Temporal Consistent Attention (TCA). TCA enforces temporal coherence through a dual-path mechanism that interpolates between standard self-attention (Path A) for frame-specific structure and a consistent path (Path B) that regularizes appearance context via Gaussian-weighted sliding window aggregation.