Key Features

Relights images and videos while preserving temporal consistency.
Jointly predicts relit RGB frames and per-frame warped environment maps.
Does not require prior camera poses for its main relighting pipeline.
Uses RGB-intrinsic fusion to preserve material appearance and illumination cues.
Supports high-resolution image relighting and 57-frame video relighting sequences.
Trains with synthetic and real-world video data for robust lighting variation.
Targets VFX, virtual production, editing, and relighting research workflows.
Provides public project resources including paper and code links.

The framework jointly predicts relit video frames and viewpoint-aligned per-frame warped environment maps. This joint formulation helps enforce geometry-light consistency without relying on prior camera poses. Relit-LiVE combines RGB-intrinsic fusion rendering, environment-map prediction, and robust multi-stage training with synthetic and real-world datasets to reduce common relighting failures such as spatial inconsistency, original-light leakage, and temporal flicker.


Relit-LiVE is useful for virtual production, video editing, VFX, scene relighting research, and data generation where lighting must be changed after capture. Its ability to infer aligned environment maps alongside the relit video makes it more useful than a simple style transfer or prompt-based relighting approach. With public code linked from the project, it is listed as a free open-source research tool.

Get more likes & reach the top of search results by adding this button on your site!

Embed button preview - Light theme
Embed button preview - Dark theme
TurboType Banner
Zero to AI Engineer Program

Zero to AI Engineer

Skip the degree. Learn real-world AI skills used by AI researchers and engineers. Get certified in 8 weeks or less. No experience required.

Subscribe to the AI Search Newsletter

Get top updates in AI to your inbox every weekend. It's free!