Generative AI
LiveAvatar
Real-time streaming avatar generation at 20 FPS. 14B parameter diffusion model deployed on 5x H800 GPUs.
About the Project
LiveAvatar is an advanced AI system developed in collaboration with Alibaba Group that generates real-time, streaming, infinitely-long interactive avatar videos driven by audio input. The system uses a 14-billion parameter diffusion model (WanS2V-14B) to create convincing animated characters that respond naturally to speech. Deployed on a 5x H800 GPU cluster, it achieves 20 FPS output for production-grade avatar experiences.
Key Features
- 14B parameter WanS2V diffusion model for photorealistic avatar generation
- Real-time streaming at 20 FPS using 5x H800 GPU cluster
- Supports 10,000+ seconds of continuous video generation
- 4-step sampling optimization with LoRA fine-tuning
- Apache 2.0 licensed with open model weights on HuggingFace
Impact
Ranked #1 on HuggingFace Papers with 1000+ GitHub stars. Enables next-generation human-AI interaction for enterprise clients.
Tech Stack
PyTorchWanS2V-14BDiffusersDeepSpeedGradioCUDA 12.4
Metrics
1000+ GitHub stars
#1 HuggingFace Paper
10,000+ sec video generation
Interested in this project?
Let's discuss how I can build something similar for you.