Manga Translator is a browser-based tool from Ficory LLC that takes scanned or digital comic pages and returns translated, relettered versions in a target language. It handles the full chain in one pass: optical character recognition finds the text, a translation model rewrites it, and smart typesetting erases the original lettering and refills each bubble. The site reports more than 100 supported languages and 100 million completed translations, and it works without any software install.
The model selector is the main quality dial. Manga, the first-generation option, costs one credit and routes Chinese, Japanese, and Korean pages to one recognition model and other languages to another, which makes the source-language setting important. Ultra costs two credits and spends more effort on fine detail, with the site noting that brief delays can occur at peak times. EagleEye also costs one credit and is described as an early-access model built on the team's newest research, aimed at users who want to try features before they are finalized.
The recognition layer is built for the messy reality of comic pages: text over busy backgrounds, lettering that curves with the artwork, and outlined or stylized fonts. After recognition, the typesetting engine erases the source text, refills the space, and adapts font size, position, and line breaks so the page keeps its original layout and reading rhythm. A one-click refill handles most pages, and the site says power users can still nudge position, font, and terminology by hand.
The font menu goes well beyond a single default. Auto picks a suitable face, while named styles map to specific storytelling needs: Comic for general dialogue and sound effects, Brush and Heavy Brush for martial arts and battle scenes, Classic and Literary for narration and period pieces, Lovely and Playful for romance and comedy, and Clean or Morden for dense webtoon dialogue. The online editor also accepts custom font files, though fonts chosen at submission time are limited to the built-in list.
The core flow is three steps: choose languages, upload files, and translate. Uploads accept jpg, jpeg, png, webp, avif, pdf, zip, tar, and tar.gz, capped at 15MB per page, and translation can start automatically after upload. Batch uploads run through a background queue so a full chapter can process while you do something else, and finished images download together. For teams that need automation, a v2 API lets developers create a key, request upload URLs, submit tasks, poll status by batch ID, and pull translated images.
Machine output rarely lands perfectly on every page, and the editor is where fixes happen. Clicking "Edit translated images" on a finished task opens an online canvas for correcting dialogue, repositioning text boxes, and adjusting fonts and layout. Exports from the editor are separate from the original task's images, so the corrected version has to be saved deliberately rather than replacing the first pass.
The service states that it uses SSL/TLS encryption in transit, access control, least privilege, audit logging, and periodic security reviews. A responsible disclosure policy asks researchers to report vulnerabilities privately before public disclosure. Support runs through email, phone, X, GitHub, Discord, and Telegram, and the company ships user-requested features on a monthly cadence.