UniLumos: Unified Image and Video Relighting
UniLumos is a unified relighting framework by Alibaba DAMO Academy for adjusting illumination in both images and videos with geometry feedback.
UniLumos
Wan2.1RelightingImage-to-VideoDAMO AcademyUnified relighting framework by Alibaba DAMO Academy for both images and videos. Brings RGB-space geometry feedback into a flow-matching backbone for physically plausible illumination adjustment.
| Developer | Alibaba DAMO Academy |
| Release Date | 2025-11 |
| Architecture | Wan2.1-based flow-matching with RGB-space geometry feedback |
| Capabilities | Image and video relighting, illumination adjustment, shadow and highlight correction |
| Speed | 20x speedup through path consistency learning |
Overview
UniLumos is a unified relighting framework developed by Alibaba DAMO Academy that handles both image and video relighting tasks within a single model. Built on a flow-matching backbone, it brings RGB-space geometry feedback to ensure physically plausible illumination adjustment with correct highlights, shadows, and occlusions.
Key features include unified handling of both images and videos, physically plausible relighting with geometry awareness, and fast processing through path consistency learning that achieves approximately 20x speedup over previous approaches. The model seamlessly blends characters into new backgrounds with automatically adjusted lighting.
ComfyUI Integration
UniLumos is available in ComfyUI via the Kijai WanVideoWrapper custom nodes. The model weights are hosted on Hugging Face under both the official Alibaba DAMO Academy repository and the Kijai WanVideo_comfy collection.
Resources
| Resource | Link |
|---|---|
| HuggingFace (Original) | Alibaba-DAMO-Academy/UniLumos |
| HuggingFace (Kijai) | Kijai/WanVideo_comfy |
| GitHub | github.com/alibaba-damo-academy/Lumos-Custom |
Guides and workflows related to this model series.