Content Authenticity Initiative Trustmark With WASI WebGPU
summary
Builds a single Wasm component that runs in wasmCloud and is transpiled for the browser to encode images and watermark video frames using the Trustmark watermark system and WASI WebGPU (wasi:webgpu). Porting the ONNX runtime requires stubbing pthreads, process IDs and exceptions; a minimal GPU-backed runtime is functional but slow for long videos, so framewise watermarking at CDN edge proves feasible to avoid sending huge files to datacenters, demo completes successfully though models and optimizations remain in progress.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:28Content Authenticity Initiative overview
- 3:36AI-generated images problem
- 5:44Three pillars of credentials
- 7:51TrustMark implementation
- 8:57WASI Onyx inference
- 10:41Edge compute for AI inference
- 12:06Challenges with OpenCV video
- 14:56Porting ONNX to WASI WebGPU
- 17:46Future of ONNX runtime
- 18:51Live demo: video watermarking
- 20:36Edge compute and WASI Cloud
- 22:01Q&A
related talks
30mWasi-Webgpu: Build Beautiful Graphics and Safely Run AI with Any GPU
33mUnleash the Power of Open Source WASM on a Hyper-Distributed Clo...
25mPathfinder to GPU Offload in WASM
31mRunning AI workloads with WASM - is it production ready? [Panel]
31mGPUs Unleashed! Make Your Games More Powerful With wasi-gfx
30mWasm Beyond the Browser: Use Cases at Scale
33mWebAssembly: The Next Frontier of Cloud Computing
08mLightning Talk: Build trusted cloud apps with Wasm: WebAssembly Micro Runtime is ready