ComfyStream

ComfyStream

Page is under construction.
Feedback Welcome!

See DeepWiki

ComfyStream is a real-time AI video processing system that enables live video transformation using ComfyUI workflows. The system captures video from a webcam or other media source, processes each frame through a ComfyUI workflow graph, and streams the processed output back with sub-second latency.

Key Capabilities:

• Real-time video processing at 15-30 FPS
• WebRTC-based streaming for low latency
• ComfyUI workflow compatibility for flexible AI pipelines
• TensorRT acceleration for 10x+ performance improvements
• Multiple deployment modes (Docker, cloud, local development)
• Bring Your Own Compute (BYOC) orchestration support

Primary Use Cases:

• Live AI video effects (style transfer, depth estimation, face animation)
• Real-time image-to-image translation on video streams
• Interactive AI art generation with webcam input
• Distributed GPU compute for video processing

Architecture

ComfyStream is organized into six primary architectural layers, each with distinct responsibilities:

Layer Responsibilities:

Layer	Components	Primary Function
Client	Browser, Webcam	Capture media input and display output
UI	StreamCanvas, Room, Settings	Video standardization, WebRTC setup, configuration
Transport	RTCPeerConnection, MediaTracks	Real-time media streaming with WebRTC
Server	app.py, byoc.py	WebRTC signaling, media track handling, orchestration
Processing	Pipeline, ComfyStreamClient	Frame-to-tensor conversion, workflow execution coordination
Backend	ComfyUI, Custom Nodes	Workflow graph execution, AI model inference

Core Components

…