PaneTrans — drag-select region translation + OCR on video/canvas, built on Transformers.js (local by default)

Great explanation. For a latency-sensitive overlay, keeping the model and WebGPU context alive in an offscreen document sounds like the right tradeoff. The hybrid approach of offscreen for inference and service worker for coordination is clean and practical. Looking forward to hearing how NLLB q8 performs in real-world use—thanks for sharing your reasoning.