This update is built from your reviews and suggestions. Thank you!
New "Clear translations automatically" (Settings > Display): translations from one-shot mode can now disappear on their own after a delay you choose, and optionally as soon as the original text vanishes from the screen. Both options are off by default.
New "Where Ollama models are stored" (Settings > Models, and in the first-run setup): local AI models can take tens of gigabytes, so you can now keep them on any drive you like. Existing models are moved safely: they are copied and verified first, and the originals are deleted only after that.
Faster drawing: rendering translations is now about 60% faster on both Full HD and 4K screens, with pixel-identical output.
The official manual is now bundled with the app. Open it from Settings > Help (Japanese and English, works offline). It is also linked from the store page sidebar.
Stability note: "High quality (AI background restoration)" now starts a few seconds after launch (after OCR warm-up) to keep the GPU work serialized.
Also fixed: translation could fail with an error ("see error log") when a local AI model returned multi-line text.