MDMartin Delophyinmartindelophy.hashnode.dev·1h ago · 9 min readRemoving Video Watermarks Locally with MI-GAN, ONNX Runtime Web, and WebGPUWatermark and unwanted-object removal looks straightforward in a model demo. Inside a real video editor, it becomes a much larger systems problem. The conventional architecture uploads media to a GPU 00
MDMartin Delophyinmartindelophy.hashnode.dev·1d ago · 6 min readBuilding In-Browser AI Music Generation: WebGPU, ONNX Optimization, and Long-Audio Stitching TricksIntroduction Running AI models entirely in the browser is no longer a futuristic experiment. With WebGPU and ONNX Runtime, we can now run diffusion-based audio models 100% locally, without any backend00
MDMartin Delophyinmartindelophy.hashnode.dev·3d ago · 1 min readWho Says Browser-Based ONNX Video Editing Has to Be Expensive or Difficult?Who says building an ONNX-powered video editing timeline in the browser has to be expensive or difficult? I vibe-coded a working prototype with AI by combining a few browser-friendly components: FFmp00
MDMartin Delophyinmartindelophy.hashnode.dev·4d ago · 10 min readWhere ffmpeg.wasm Fits — and Doesn’t — in a Browser-Based Video EditorBuilding a video editor that runs entirely in the browser changes how you think about media processing. The real implementation is open source if you want to inspect the export and fallback paths: htt00
