Writing
Technical write-ups and architectural breakdowns from real engineering projects.
2026-08-248 min read
Running LaMa Inpainting in the Browser — WebGPU First, WASM Where It Counts
How Inori Studio runs LaMa inpainting entirely on-device: where ONNX Runtime Web + WebGPU worked, where it broke, and why the answer was a hybrid WebGPU/WASM pipeline.
Local AIWebGPUONNXInori Studio
2026-08-247 min read
Anatomy of an Automated Manga Translation Pipeline
Building a Python CLI that takes raw manga chapters to typeset translations: bubble detection with morphological dilation, dominant-color text cleaning, PaddleOCR, and Pillow-based typesetting.
PythonOpenCVOCRAutomation