Alibaba Quark Open-Sources LiveAvatar: 14B Model for Real-Time Infinite Audio-Driven Avatars
tom_doerr · x · 2026-08-12
Alibaba's Quark team open-sourced LiveAvatar, a project accepted as an Oral presentation at ECCV 2026.
- Core Capabilities: Utilizes a 14B parameter diffusion model to generate avatars driven by real-time audio.
- Technical Highlights: Employs an algorithm-system co-designed framework enabling streaming and infinite-length continuous generation.
- Open Source: The project has garnered 2.3k stars on GitHub, providing complete inference code and multi-GPU deployment scripts.
More from Multimodal
- AI Video Generation Bottleneck: Full Regens Kill Iteration and Momentum — SucceededMind · 2026-08-12
- LTX-2.5 Live in ComfyUI: Introduces Dynamic Compute Budget Rendering — Comfy-Org · 2026-08-12
- Midjourney Officially Rolls Out Version 8.2 Update — azed_ai · 2026-08-12
- Boost LTX 2.5 Video Generation Speeds by 5x with Conv VAE Swap — desktop4070 · 2026-08-12
- Dev Builds macOS Local Image App 'Diffusion Bus' Using Claude Code — deckarep · 2026-08-12
- Flux 3 [T2V] Demo: Generating 1990s-Style Monster Footage — CurieuxExplorer · 2026-08-12