LiveAnimate achieves real-time, stable long-form human animation

Researchers have developed LiveAnimate, a novel system capable of generating stable, long-form human animation in real-time. Built on a 14-billion parameter video Diffusion Transformer, the system employs a two-stage training pipeline and a unique Pose-Retrieval Sink Attention mechanism to maintain consistent appearance and identity over extended durations. This allows for interactive applications like live streaming and virtual avatars, achieving 19.63 FPS inference on high-end GPUs and demonstrating minimal degradation in quality over a three-minute benchmark. AI

IMPACT Enables real-time interactive applications like live streaming and virtual avatars by overcoming latency and stability issues in diffusion-based animation.

RANK_REASON The item describes a new research paper detailing a novel system for real-time human animation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Block-wise Self-Forcing Distillation
• Diffusion Transformer
• LiveAnimate
• NVIDIA H100
• Pose-Retrieval Sink Attention
• Reference-Anchored Teacher-Forcing Adaptation
• Static Sink
• Ulysses

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →