Researchers have developed Hallo-Live, a novel framework for real-time text-driven audio-visual avatar generation. This system utilizes an asynchronous dual-stream diffusion approach combined with human-centric preference distillation to achieve high fidelity and synchronization. Hallo-Live demonstrates significant speed improvements, running at 20.38 FPS with low latency, making it suitable for interactive applications. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Enables real-time, interactive avatar generation for applications like virtual assistants and streaming.
RANK_REASON This is a research paper detailing a new framework for real-time avatar generation.