Light Control in Photos Now Works Like a 3D Scene

Reshaping the light in a photo after the shot has usually meant either giving up or spending hours on it. A study presented for SIGGRAPH 2025 aims to make this feel more like adjusting a light panel in 3D software. The method, by Chris Careaga and Yağız Aksoy, allows users to insert spot, point or environmental lights into a scene, according to the report.

The system infers a colored 3D representation of the scene from a single photograph. That representation is used to compute where light falls; a path-tracing engine then renders an approximate version of the scene. In the final step, a neural renderer turns this rough image into a photorealistic result.

According to the paper, the team developed a differentiable rendering process to reconstruct in-the-wild illumination, enabling self-supervised training on raw image collections. This lets the network learn from everyday photos without pre-known scene data.

The approach is presented as a significant step toward bringing the explicit light control found in tools like Blender to photographs. In practice, it could let someone redirect light in a product shot or try different lighting scenarios on an interior image. The methodology is protected under intellectual property; licensing inquiries are directed to SFU Technology Licensing Office and Dr. Yağız Aksoy.