FlashRelight: Portrait Video Relighting with Dynamic Lighting
Abstract
Portrait video relighting aims to modify the illumination of a portrait video while preserving photorealism and temporal stability. The task becomes particularly challenging under dynamic lighting, where illumination direction, intensity, and color vary over time. To tackle this problem, we introduce FlashRelight, a novel two-stage framework for dynamic portrait video relighting. FlashRelight first suppresses the source illumination through delighting, and then synthesizes the target dynamic lighting on an illumination-neutral portrait video. FlashRelight supports three complementary control modalities: envmap sequences for explicit frame-level lighting control, dynamic lighting text for semantic illumination editing, and lighting reference videos for visual lighting transfer. Training follows a low-cost hybrid strategy that combines static one-light-at-a-time (OLAT) data with generated dynamic-lighting portrait videos, covering both controlled illumination variation and realistic portrait motion. Extensive experiments show that FlashRelight achieves temporally coherent and photorealistic dynamic relighting while enabling flexible cross-video lighting transfer. Code is available at https://github.com/vivoCameraResearch/FlashRelight
The following analysis is generated by AI and may contain errors. It is separate from the authors’ abstract and is not peer review.
Research summary
FlashRelight introduces a robust two-stage framework for dynamic portrait video relighting that supports multiple control modalities. By combining delighting with dynamic lighting synthesis and leveraging a hybrid training strategy, the method achieves state-of-the-art temporal coherence and photorealism, representing a significant step forward in video editing technologies.
Methodology
The paper proposes FlashRelight, a two-stage framework for dynamic portrait video relighting. The core methodology involves first suppressing the source illumination (delighting) to create an illumination-neutral video, followed by synthesizing target dynamic lighting. The novelty lies in supporting three complementary control modalities: environment map sequences, dynamic lighting text, and lighting reference videos. The training strategy is a hybrid of static one-light-at-a-time (OLAT) data and generated dynamic-lighting videos. This approach addresses the temporal stability challenge inherent in video relighting, which is a significant improvement over frame-independent methods.
Results & evaluation
The experiments demonstrate temporally coherent and photorealistic results under dynamic lighting conditions. The ability to perform cross-video lighting transfer is a strong practical contribution. While the abstract claims extensive experiments, the provided text is a skeleton, so the depth of quantitative metrics (e.g., FID, LPIPS, temporal consistency scores) compared to state-of-the-art baselines like ReLight or PortraitLight cannot be fully verified from the snippet alone, but the acceptance at NeurIPS 2026 suggests rigorous validation.
Reproducibility
The authors have released code at a GitHub repository (vivoCameraResearch/FlashRelight), which significantly enhances reproducibility. The use of standard control modalities (envmaps, text) also aids in benchmarking against other methods.
Limitations
The reliance on generated dynamic-lighting portrait videos for training may introduce artifacts if the generation model is imperfect. The two-stage process (delight then light) may accumulate errors. The paper focuses on portraits, limiting generalizability to other object categories.
Broader impact
This work has high practical impact for video editing, virtual production, and AR/VR applications where dynamic lighting is crucial for realism. It enables creators to modify lighting in post-production with high fidelity, a capability previously limited to static images or low-quality video. FlashRelight introduces a robust two-stage framework for dynamic portrait video relighting that supports multiple control modalities. By combining delighting with dynamic lighting synthesis and leveraging a hybrid training strategy, the method achieves state-of-the-art temporal coherence and photorealism, representing a significant step forward in video editing technologies.
Author of this paper? Send a correction.