TeleFuser¶

A high-performance runtime for world model inference and multimodal generation.

Features¶

🌍 World Model Runtime — Continuous execution, stateful sessions, bidirectional control loops
🚀 High Performance — Optimized Triton kernels, feature caching, and multi-GPU inference
🎨 Multimodal Generation — Image/video generation, super-resolution, speech-to-video
📡 Streaming Transport — WebRTC with media tracks plus DataChannel for real-time inference
🔧 Flexible Configuration — Attention implementations, parallel strategies, quantization, offloading
📦 Extensible — Easy to add new models, stages, and pipelines

Supported Models¶

World Model and Real-Time¶

Model	Tasks	Description
LingBot-World-Fast	Bidirectional streaming	Interactive world model via WebRTC DataChannel

Video Generation¶

Model	Tasks	Description
WanVideo (Wan2.½.2)	T2V, I2V, FL2V	Video generation and editing
HunyuanVideo	T2V, I2V	Video generation
LTX Video	I2V + Audio	Video generation with audio
FlashVSR	VSR	Video super-resolution
LiveAct	S2V	Speech-to-video
LongCat-Video	T2V, I2V	Long video generation

Image Generation¶

Model	Tasks	Description
Qwen-Image	T2I, Edit	Image generation and editing
Z-Image	T2I	Image generation
Flux2 Klein	T2I	Image generation

Quick Start¶

# Install
pip install telefuser

# Batch serving
telefuser serve /path/to/pipeline.py --port 8000

# Stream serving (requires WebRTC)
pip install -e ".[webrtc]"
telefuser stream-serve examples/stream_server/stream_lingbot_world_fast.py -p 8088

Documentation Sections¶

Service Guide — Batch serving, task APIs, and SDK
Stream Server — WebRTC streaming and bidirectional control
Configuration — Runtime and model configuration
Parallel Inference — Distributed processing strategies
Adding New Model — Integrate new models
Profiler — Performance analysis tools

Switch to Chinese 🇨🇳