VideoCompositor: a Case Study of Live Video Composing
summary
VideoCompositor is a live video compositor that composes RTP streams into a single output with custom shaders, text and CSS-like overlays while targeting sub-50 milliseconds latency for conferencing and higher for HLS. It prototypes pipelines using FFmpeg for I/O, moves heavy work to the GPU via WebGPU/wgpu (with plans for Vulkan/DPU decode/encode), exposes a language-agnostic RTP + HTTP control API, and shows a working grid demo but warns that browser rendering forces costly GPU↔CPU memory copies that hurt scale.
AI-written summary · about
chapters (8)
- 0:00Introduction to VideoCompositor
- 1:52Key challenges in video composing
- 3:41Exploring graphical APIs
- 6:37How VideoCompositor works
- 7:19Live demo of the tool
- 8:05Custom shaders in VideoCompositor
- 9:11Web integration with Chrome engine
- 10:39Closing and Q&A
related talks
21mDynamic video in the browser
19mPuppeteer + canvas + WebCodecs to replace
29mStop Fighting Hydra - Replacing Headless Chromium
18mVideostrates: Collaborative, Distributed and Programmable Video Manipulation
21mReal-Time Video Super Resolution for Live Streaming with WebGPU
12mExploring Video Editing in the Browser
49mVoctomix
16mDemocratizing live video with React