🎨 Video Restyling with Full Identity Preservation
The ID-V2V framework has been introduced, which allows for changing the environment, lighting, and style of a video while strictly preserving the character's facial features, micro-expressions, and lip synchronization. The technology is based on the Wan2.1, SAM3, and DiT architectures.
🌍 ID-V2V solves the problem of the shortage of paired data for training video models by creating training pairs from single videos through a relighting inversion technique. This paves the way for professional use of neural networks in production.
👤 The tool allows "transferring" a person into any setting without turning them into a digital twin. This significantly increases the controllability of generative video.
Source 1: https://eyeline-labs.github.io/ID-V2V/ Source 2: https://arxiv.org/abs/2607.22830