v2.0

First release of this fork: TextGrab on steroids. Capture any screen with one gesture, then select, copy and translate its text. Recognition runs on your device and needs no Google Play services.

Highlights

• Instant capture through Shizuku: the quick settings tile, the assistant gesture and an adb command grab the current screen directly, without saving a screenshot.
• Assistant gesture: long-press home or the navigation handle to grab text.
• Capture preview: the screen opens as an inset card over a blurred backdrop, so text at the edges is easy to reach.
• Native-feeling selection: dragging follows lines, handles keep their grab offset, copied text has no stray empty lines.
• Translation in place: every line is replaced by its translation on the image, and the translated text is selectable too.
• Five translation engines: on-device ML Kit (default, offline), DeepL, Google Cloud Translation, any OpenAI-compatible LLM endpoint, or Claude, each with your own API key.
• Language pack manager: download packs ahead of time, delete the ones you don't need.
• Better recognition: icons and stray symbols are filtered out, lines are ordered as on screen, and Chinese, Japanese and Korean are bundled.
• Redesigned Material 3 interface with a guided setup on the home screen.

Getting started

1. Install the APK and open TextGrab.
2. Tap Set up Shizuku and allow access.
3. Add the OCR screenshot tile to quick settings.
4. Optional: enable Keep TextGrab ready and set TextGrab as the
   assistant app.

Notes

• The APK is about 110 MB; most of it is the four bundled recognition models.