ControlLight: Towards Controllable, Consistent, and Generalizable Low-Light Enhancement

ControlLight is a controllable low-light enhancement framework for improving real-world dark images.

It is trained on a large real-world dataset with continuous illumination-strength supervision.

A misalignment-aware weighted flow matching loss helps preserve consistency across different enhancement levels.

The method improves controllability, generalization, and visual quality, and reports state-of-the-art results.