🛠 HOMIE: Video Personalization via Human-Object Interaction
The HOMIE framework, based on the Wan2.1-T2V-14B architecture, has been introduced. The system leverages the capabilities of MLLMs to generate videos with high precision in preserving the identity of characters and objects (HOCVP).
🌍 The technology addresses the problem of loss of detail and inconsistency during generation, opening up possibilities for professional AI use in advertising and branded content.
👤 Users will be able to create realistic clips where specific people or branded products maintain their features and exact logo appearance under any camera movements.
Source 1: https://arxiv.org/abs/2607.18217 Source 2: https://yiyangcai.github.io/homie-page.github.io/