🎬Generative video is rapidly becoming a foundational capability for content creation, but professional filmmaking VFX requires far more than visually compelling generation. Models must offer precise control over what changes, and reliably preserve everything else. That’s why we are introducing 🌟ID-V2V🌟 at @SIGGRAPHAsia 2026!
🖌️Precise control: Redesign the environment, lighting, and visual style of an existing video in post-production using edited keyframes.
🎥Reliable preservation: Treats identity preservation as a video relighting problem, keeping facial likeness, subtle expressions, eye gaze, and lip-sync completely intact.
📚Data pipeline: Solves the lack of paired training data and seamlessly supports both single- and multi-subject scenarios.
Congrats to @Yuancheng_Xu0, Mingming He, @pablosalamancal, @maleewahaha, @yash2kant, Emmett Steven, @debfx, and @realNingYu from @netflix and @eyelinestudios.
#AI #AIResearch #MachineLearning #AIGC #GenAI #VideoGeneration
🖌️Precise control: Redesign the environment, lighting, and visual style of an existing video in post-production using edited keyframes.
🎥Reliable preservation: Treats identity preservation as a video relighting problem, keeping facial likeness, subtle expressions, eye gaze, and lip-sync completely intact.
📚Data pipeline: Solves the lack of paired training data and seamlessly supports both single- and multi-subject scenarios.
Congrats to @Yuancheng_Xu0, Mingming He, @pablosalamancal, @maleewahaha, @yash2kant, Emmett Steven, @debfx, and @realNingYu from @netflix and @eyelinestudios.
#AI #AIResearch #MachineLearning #AIGC #GenAI #VideoGeneration
🎬Capture the performance first and redesign the look later!
🚀Introducing ID-V2V, @netflix latest research exploration in human-centric video editing, to appear at @SIGGRAPHAsia 2026. It enables a powerful creative workflow: creators can redesign the environment, lighting, and visual style of an existing video after capture while preserving the original human identity and performance.
🧩Built for production workflows, ID-V2V introduces identity-preserving video restylization. Given a source video and edited keyframes, ID-V2V propagates those edits across the entire video while preserving human appearance, subtle facial expressions, full-body motion, and multi-person interactions. This gives creators the freedom to reshape the visual world in post-production while keeping the original performance intact.
Kudos to the team: @Yuancheng_Xu0 , Mingming He, @pablosalamancal , @maleewahaha , @yash2kant , Emmett Steven, @debfx , @realNingYu .
💻 Code: github.com/Eyeline-Labs/I…
🎬 Project Page: eyeline-labs.github.io/ID-V2V/
📄 Paper: arxiv.org/abs/2607.22830View quoted post on X
