Reproduce “TokenLight: Precise Lighting Control in Images using Attribute Tokens” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sumit Chaturvedi, Yannick Hold-Geoffroy, Mengwei Ren, Jingyuan Liu, He Zhang, Yiqun Mei, Julie Dorsey, Zhixin Shu
A novel method introduces precise and continuous lighting control in 2D images by representing illumination changes as attribute tokens within a diffusion transformer. This approach achieves physically plausible relighting effects and outperforms existing methods in quality and accuracy without requiring explicit 3D scene reconstruction.
Key insights
- A unified, compact lighting-attribute token representation enables disentangled and continuous control over diverse physical lighting factors within a single diffusion architecture.
- The model implicitly learns an inherent, geometry-aware understanding of light-scene interactions from synthetic data, allowing physically plausible effects without explicit 3D scene reconstruction.
- Conditioning on canonical 3D light parameters, coupled with scene-agnostic similarity transformations during training, facilitates interactive 3D light control consistent across various viewpoints.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.