Relit-LiVE Enables Video Relighting via Joint Environment Video Learning
TL;DR
Relit-LiVE repurposes large-scale video diffusion models as neural renderers. It decomposes videos into intrinsic scenes and renders them under novel illumination.
What changed
Relit-LiVE introduces video relighting through joint learning of environment videos using large-scale video diffusion models. It decomposes input videos into intrinsic scene parts and renders under novel lights without needing exact decomposition. This merges decomposition and rendering into one diffusion framework.
Why it matters
Relit-LiVE beats Stable Video Diffusion baselines by 18 percent on LPIPS scores for dynamic relighting on the DAVIS dataset. Developers gain a plug-and-play module for video editing pipelines handling real-world motion. Basic users access editable lighting in diffusion workflows.
What to watch for
Compare Relit-LiVE outputs to 3D Gaussian Splatting relighters on temporal flicker in the HyperNeRF dataset. Pull the model from Hugging Face and relight a 5-second clip with point lights to measure consistency via frame-by-frame PSNR. Check repo issues for fixes on high-resolution inputs.
Who this matters for
- Vibe Builders: Apply cinematic lighting shifts to existing video footage to create mood-consistent visual narratives.
Harsh’s take
Relit-LiVE attempts to solve the persistent headache of video relighting by folding decomposition and rendering into a single diffusion process. By bypassing the need for perfect scene geometry, it offers a pragmatic path for creators who need lighting control without the overhead of full 3D reconstruction. The 18 percent improvement over existing baselines suggests the approach has legs for production workflows.
However, the real test remains temporal stability. Diffusion models often struggle with frame-to-frame coherence, and relighting adds a layer of complexity that frequently results in flickering artifacts. Until the model proves it can handle high-resolution inputs without breaking under motion, it remains a promising research artifact rather than a production-ready tool.
Operators should prioritize testing consistency on complex motion sequences before integrating this into any serious video pipeline.
by Harsh Desai
More AI news
- Daily RoundupVercel Flags and WebSockets, Google Interactions API, and agent tools for live apps
Vendors released feature flags, WebSocket support, unified model APIs, new video models, trending OCR tools, and agent deployment options on 22 June, giving builders direct paths to ship realtime and segmented AI features.
- FeatureLovable Build with URL links now reference public web pages
Lovable's Build with URL links can now reference public web pages alongside images. The feature uses the referenced page's layout, content, and styling to recreate or iterate on it.
- FeatureSet up cloud environments and run subagents with /in-cloud
Cursor's /in-cloud sets up cloud development environments in under 10 minutes and runs isolated subagents. Sessions hand off between local machines and the cloud.