Hi-Light: A Path to high-fidelity, high-resolution video relighting with a Novel Evaluation Paradigm

TL;DR

This work tackles video relighting by addressing flicker and detail loss without training new models. It introduces Hi-Light, a training-free pipeline that combines a lightness-prior anchored diffusion process, the Hybrid Motion-Adaptive Lighting Smoothing Filter, and LAB-DF to stabilize illumination and preserve high-frequency details in high-resolution video. To evaluate lighting quality specifically, the authors propose the Light Stability Score, together with SSIM for detail fidelity, and demonstrate state-of-the-art performance against baselines on diverse footage. The approach yields stable, detail-rich relit videos and offers a robust, scalable framework for lighting-aware video editing, with broad practical implications for content creation and professional workflows.

Abstract

Video relighting offers immense creative potential and commercial value but is hindered by challenges, including the absence of an adequate evaluation metric, severe light flickering, and the degradation of fine-grained details during editing. To overcome these challenges, we introduce Hi-Light, a novel, training-free framework for high-fidelity, high-resolution, robust video relighting. Our approach introduces three technical innovations: lightness prior anchored guided relighting diffusion that stabilises intermediate relit video, a Hybrid Motion-Adaptive Lighting Smoothing Filter that leverages optical flow to ensure temporal stability without introducing motion blur, and a LAB-based Detail Fusion module that preserves high-frequency detail information from the original video. Furthermore, to address the critical gap in evaluation, we propose the Light Stability Score, the first quantitative metric designed to specifically measure lighting consistency. Extensive experiments demonstrate that Hi-Light significantly outperforms state-of-the-art methods in both qualitative and quantitative comparisons, producing stable, highly detailed relit videos.

Figures (19)

• Figure 1: Demonstrations of the text-conditioned video relighting task by our framework.
• Figure 2: Comparison of the relighting effects of CapCut Sunset filters and our model.
• Figure 3: The overall structure of our Hi-Light framework. The framework first processes a downsampled video through a guided relighting diffusion loop to generate lighting information where a lightness prior is anchored. The intermediate output is then stabilized using an HMA-LSF to eliminate flickering. Finally, the LAB-DF module transfers the illumination information to the high-resolution source.
• Figure 4: LAB feature maps will be extracted from the intermediate relit video and the input video. The high-frequency information from the input video will be fused with the illumination information from the intermediate relit video.
• Figure 5: A visual comparison of relighting methods with the text prompt “sunset lighting.” Hi-Light achieves the best detail preservation. The top row shows a relit video frame, while the dotted red box marks regions of contrast, enlarged in the bottom row.