Title screens, “New Game / Continue” buttons, HUD labels, the sign above a shop — no extractor can reach them, because they are pixels rather than strings. The Image Studio reads that text, rebuilds the artwork underneath it, and typesets your translation back into the picture.
The workflow
The same four steps a professional comic translator works through — and every one of them is yours to correct, rather than a single automatic pass you either accept or abandon.
One click finds every text area on the image and reads it. The lines of one speech balloon are grouped into a single block, so you translate a sentence instead of three fragments of one. Anything it misses you can box by hand, and any single box can be re-read on its own without re-running the page.
Paint over the original lettering with a brush or a lasso and the art underneath is reconstructed — a real background, not a coloured patch stamped over it. The eraser takes area back out of your selection, so a mask that grabbed too much is a correction rather than a restart.
Font, size or auto-fit, weight, alignment, line height, letter spacing, padding, outline, shadow and colour — per text box. Vertical Japanese sets in proper columns with the standard line-breaking rules, and an eyedropper lifts a colour straight off the artwork so your caption matches the ones around it.
On export the translated image is written under its original filename, in its original format, so the game loads it with no patching. The untouched original is kept alongside it, so you can always go back.
What the editor gives you
Automatic passes are a starting point. Everything below exists because the automatic answer is sometimes wrong, and being unable to fix it is what makes a tool useless on real artwork.
Draw, move, resize, rotate and delete text boxes, select several at once, lock or hide the ones you have finished. Undo and redo throughout, so nothing you try is a commitment.
Brush and lasso choose the area to rebuild; the eraser takes it back. The result appears on the canvas immediately instead of at export, so a bad mask is something you see and fix, not something you discover in game.
Auto-fit sizes text to its box, wrapping where a line should break. Japanese, Chinese and Korean wrap per character with proper kinsoku rules, so a line never opens with a closing bracket or a full stop.
Seven bundled faces cover Latin, Greek, Cyrillic, Japanese, Simplified and Traditional Chinese, Korean, Thai and Arabic — so a translated title renders as words instead of a row of empty boxes.
The canvas lays text out with the same engine the export uses, so the preview cannot disagree with the result. Move a box and the text re-wraps exactly as it will in the finished image.
Recognized text goes through the provider you already use for dialogue — DeepL, Google, OpenAI, Claude, DeepSeek or a local model — and lands in the editor where you can rewrite it to fit the space.
Engine support
Image text is found wherever a game keeps its pictures, so the reach is wide. These are the engines with dedicated handling, including the archives and encryption they ship with.
Title screens, menus, pictures and system art. Games that ship their images encrypted (.rpgmvp / .png_) are read and re-encrypted on the way out, so the engine still loads them.
Art kept loose under `game/`, and art packed inside `.rpa` archives — read straight out of the archive and written back as a loose file the engine loads first, with no repacking.
Images inside XP3 archives, TLG format included, delivered as a patch archive the game mounts by itself — no repacking of the original data.
Images inside .nsa and .sar archives, written back as loose files, which the engine reads in preference to the archive.
Unity, Wolf RPG, TyranoBuilder, Godot and the rest: any game that keeps its images in ordinary folders is scanned and can be translated the same way.
Plans
Reading your own work is free. The editor runs on a hosted recognizer and inpainting service, which is what the paid plans pay for.
Open the Studio on any project, pan and zoom, and read every translation you have already made. Changing images needs a plan.
The entire editor: detection, reading text, cleanup, typesetting and baking, with a daily allowance sized for ordinary work.
Everything in Supporter with a much larger daily image allowance, for translating a whole game's artwork rather than a few screens.
Detection, reading and cleanup all run on our own hosted service, so each plan comes with a daily allowance. It resets at midnight UTC.
| Operation | Free | Supporter | Pro |
|---|---|---|---|
Detect text One pass over a whole image: finds every text area on it and reads them all at once. | — | 100 | 500 |
Read one box Reads the text inside a single box you framed — for fixing a box detection got wrong, or one you drew yourself. | — | 500 | 2,500 |
Clean an area Erases the original lettering and rebuilds the artwork behind it. | — | 250 | 1,250 |
Image area The total picture area processed, counted in megapixels. A full 1920×1080 page is about 2 MP; a menu button is a fraction of one. | — | 120 MP | 600 MP |
In practice that is roughly 50 full-size pages a day on Supporter and 270 on Pro. Small menu and button images cost a fraction of a page, so they go much further.
Every figure is per day. Image area is usually what runs out first on full-page work, and almost never matters when you are translating small buttons. The free plan can open the editor and read translations you already made, but does not process images.
Because it is not text. A title screen is a picture, and the words on it are coloured pixels baked into that picture when the artist drew it. There is no string in the game's data to replace, so the only way to translate it is to read the image, repair the artwork, and draw new words into it.
On flat or lightly textured backgrounds — the usual case for menus and buttons — the rebuilt area is hard to spot. Over busy artwork it is harder, which is why cleanup is a brush you control rather than an automatic pass: you decide exactly what gets rebuilt and can correct it until it looks right.
Yes. Recognition follows your project's source language, and the bundled fonts cover Latin, Greek, Cyrillic, Japanese, Chinese, Korean, Thai and Arabic on the target side.
No. Translated images are written into your export folder, and the untouched original is kept next to each one, so you can revert at any point.
The area being read or cleaned is sent to the recognizer for that operation and is processed in memory — it is not stored. Everything else, including your translations and typesetting, stays on your machine.
Get started
The Image Studio ships in RuneTranslate for Windows. Install it, open a project, and the artwork is a tab away.
