Manga OCR that reads the bubbles and translates them

What manga OCR has to deal with

• Vertical lines that run top to bottom and right to left.
• Hand-lettered and stylised fonts, bold shouts and tiny whispers.
• Text on top of screentone, speed lines and busy art.
• Pages with many bubbles, in a reading order that jumps across panels.

How RuneManga reads a page

1. It finds the speech bubbles, captions and text on the art.
2. It reads each one with the recognizer for its script: Japanese, Korean, Chinese, or the Latin alphabet for English, Spanish, French and more.
3. It translates the lines with RuneLingua, RuneManga's own AI translator for manga and manhwa.
4. It erases the original text, fills in the background and letters the translation in.

Languages it reads

Japanese (vertical and horizontal), Korean, Chinese (Simplified and Traditional), English, Spanish, French, Portuguese, Indonesian, Vietnamese and Thai. The language is detected by itself, and the translation can go into 93 languages.

See the text it read

In the WebApp's editor, the list of a page's bubbles shows the text RuneManga read from the page next to its translation, so you can check a line against the source and change the translation by hand.

What it does not do

• Sound effects drawn into the art are left alone, on purpose.
• It is not a text export tool: the result is a translated page, not a text file.
• Very small or blurry text can be missed, and a sharper scan helps.

Questions

• OCR (optical character recognition)
• How does AI manga translation work?
• Why RuneManga misses some text on a page
• Raw manga translator: read Japanese raws in your language