How do you mine vocabulary from both web pages and video hardsubs?
YuzuLingo · Nov 1, 2026
Mine both surfaces with one dictionary stack: scan real page text under the pointer, and OCR burned-in video lines into selectable overlays. From the same popup you export the expression, reading, gloss, sentence context, and a screenshot into Anki or another SRS.
What immersion mining needs
A useful card pairs a headword with meaning, reading, and the sentence you met it in. Optional fields add pitch, frequency, audio, and a still from the moment. The bottleneck is getting reliable text from whatever you are consuming that day.
Web pages and other DOM text
Articles, forums, and many manga viewers already expose characters in the document. A page scanner resolves the caret under the pointer, detects the learning language among your packs, deinflects, and opens the longest match. No OCR step is required when the text is already selectable.
Video hardsubs as the other half
Hardcoded anime and drama lines need a crop, a line-change detector, and on-device recognition before mining can start. The recognized string becomes overlay text you can hover like a caption track. Softsubs and site transcripts skip OCR when the player already provides a string.
One export path
Whether the hit came from a blog paragraph or an OCR overlay, the popup can send fields through AnkiConnect in one click. Keeping page scan and video lookup on the same dictionaries avoids rebuilding decks with mismatched gloss styles.
How this differs from soft-sub-only tools
asbplayer, Migaku, and Language Reactor shine when a subtitle track or ASR transcript exists. They do not read hardsub pixels. Pairing DOM scan with OCR covers the days you read on the web and the nights you watch streams that only burn text into the frame.
