Download KotobaSync
For Macs with macOS 15 or later and Apple silicon (M1 or later). macOS 14 and earlier and Intel Macs are not supported. App interface: Japanese.
Publishing, distributing or selling subtitles you made for videos whose copyright belongs to someone else — films, series, anime, streaming titles, TV programmes and so on — or videos with those subtitles added, without the rights holder’s permission may infringe copyright law. Do not use the app that way, and do not use it commercially. KotobaSync is a tool for adding Japanese subtitles to videos such as your own recordings or recordings of international conferences and talks, to watch them yourself. The author accepts no liability whatsoever for use that breaks these terms. Read the Terms of Use
Latest version 1.2.75
Released November 10, 2026. A DMG file (14.3 MB).
Agree to the Terms and Download
By downloading, you agree to the Terms of Use and the Privacy Policy & Disclaimer. The app also asks you to agree to the Terms of Use the first time you open it.
How to verify the downloaded file
Run these two lines in the Mac’s Terminal to check that the file is the official one (if the file is in your Downloads folder).
spctl -a -vv -t open --context context:primary-signature ~/Downloads/KotobaSync-1.2.75.dmg shasum -a 256 ~/Downloads/KotobaSync-1.2.75.dmg
- If the first line prints “accepted” and “source=Notarized Developer ID”, the file has been notarized by Apple.
- If the value from the second line matches the SHA-256 in the table above, the file is unchanged since it was published.
Notarization is Apple’s automated check for known malware. It does not guarantee the app’s quality or behaviour.
How to install
The first time you open it, the Welcome to KotobaSync (ことばシンクへようこそ) screen appears and lists what to prepare before use.
Trying the tiny model first, to see that it works, is a good idea too.

-
Open the DMG
Double-click the downloaded “KotobaSync-1.2.75.dmg”.
-
Drag it to the Applications folder
Drag “ことばシンク” (KotobaSync) onto “Applications” next to it.
-
Open the app
Open “ことばシンク” in the Applications folder. The first time, macOS may show a screen confirming that it is “an application downloaded from the internet”. If you downloaded it from this page, choose “Open”. A small “About this app” (このアプリについて) window also appears once.
-
Finish the preparation
On the Welcome to KotobaSync (ことばシンクへようこそ) screen, ① download a transcription model and ② choose and set up a translation engine (only if you will translate). ③ ffmpeg is needed only to open MKV and similar files. How to prepare each one is explained in the User Guide.
If you use Homebrew, you can also install it by running the following in Terminal. It downloads the same file as the official site and checks its SHA-256 before installing.
brew install --cask kenji-miygauchi/apps/kotobasync
Requirements
| macOS | macOS 15 Sequoia or later |
|---|---|
| Mac | Apple silicon (M1 or later). Intel Macs are not supported |
| Disk space | About 16 MB for the app. A transcription model is needed separately (the default large-v3-turbo is about 1.6 GB, tiny is about 75 MB). Converting MKV and similar files needs about as much free space as the original video |
| Translation engine | Needed only for translation and review. Ollama or LM Studio (run on this Mac; you need to download a model, and the more memory you have, the larger the model you can use), or Codex (you need to sign in to ChatGPT; the subtitle text is sent to OpenAI) |
| ffmpeg | Needed only to open MKV, WebM and similar files, and to export videos with subtitle tracks (brew install ffmpeg) |
| Permissions | Usually none. Speech Recognition permission is asked for only if you choose Apple Speech as the transcription method |
| Network | Only four: downloading the transcription model (Hugging Face), translation (OpenAI only if you choose Codex; Ollama and LM Studio stay on the Mac), looking up title information (after asking you; Wikipedia and Wikidata, and TMDb only if you enter an API key), and the update check (k386.sub.jp; can be turned off) |
| Cost | Free. No account registration is needed |
| App interface | Japanese (on this page, Japanese labels are shown in parentheses) |
How to update
From 1.2.1, the app updates to new versions without you doing anything. So as not to interrupt your subtitle work, it never restarts while you are working.
- Once a day, the app compares the version information on the official site with the version you have.
- If there is a new version, it downloads it in the background and checks the SHA-256, the Developer ID signature and Apple’s notarization. If these cannot be confirmed, nothing is changed.
- The app is replaced the next time you quit it, and the next time you open it you are told, for example, “Updated from 1.2.15 to 1.2.16” (1.2.15 から 1.2.16 に更新しました).
To update right away, choose “Check for Updates…” (アップデートを確認…) in the app menu (“ことばシンク” at the top left of the screen) and press “Restart and Update Now” (今すぐ再起動して更新). If you have unsaved subtitles, it first asks whether to save them. The automatic check and the automatic installation can each be turned off in the same menu with “Check for Updates Automatically” (アップデートを自動で確認) and “Install Updates Automatically” (アップデートを自動でインストール). If you turn off automatic installation, you only get a notice when a new version is found. If the app is opened from somewhere other than the Applications folder (such as inside the disk image), or if you don’t have permission to replace it, it cannot update itself. In that case, install the new DMG from this page.
If you use 1.2.0, a notice saying “ことばシンク 1.2.75 is available” (ことばシンク 1.2.75 が公開されています) appears when a new version is found. Press “Update and Relaunch” (更新して再起動) to switch to the new version; after that it updates automatically.
New versions are also announced on this page and on Updates (also available via RSS). Models, settings, glossaries and so on are kept as they are.
How to uninstall
- Quit KotobaSync.
- Move “ことばシンク” from the Applications folder to the Trash.
- Transcription models, autosaves, saved translation progress, glossaries and work notes are in
~/Library/Application Support/SubtitleStudio. If you don’t need them, delete the whole folder (the models can take up several GB). - If you don’t need them, also delete the temporary location for converted videos,
~/Library/Caches/SubtitleStudio, and the settings file~/Library/Preferences/com.subtitle.studio.plist. - If you entered an LM Studio API token or a TMDb API key, delete the
com.subtitle.studio.lmstudioandcom.subtitle.studio.tmdbitems in Keychain Access.
Saved SRT files, converted MP4 files and exported M4V files are in the same folder as the video (or in the folder you chose in Settings). They remain even after you delete the app.
Release notes
1.2.75 (build 115) — November 10, 2026
- When reading subtitles burned into the picture, the app now tells apart on-screen text that is not subtitles (signs, car number plates, logos, titles that appear bit by bit, end credits and so on) by the size, horizontal position, colour and movement of the text. It learns the shape of the subtitles in that video and automatically removes rows that are clearly different after reading (right afterwards, “Restore removed rows” (消した行を戻す) brings them back). Rows it is unsure about are not removed but listed under “Check reading” (読み取り要確認) with the reason. Text shown near the subtitles also gets mixed into subtitle rows less often.
- Added a further measure against macOS text recognition stopping responding during reading. Reading now uses the CPU from the start, and in a test reading a 74-minute video from start to finish it no longer stopped (reading speed is unchanged).
1.2.74 (build 114) — November 9, 2026
- Fixed macOS text recognition sometimes stopping responding while reading subtitles burned into the picture, leaving the process stuck at the same point. If reading one frame takes 30 seconds or more, it is cut off, read again on the CPU, and the process moves on. If it still cannot be read, a notice appears instead of the process staying stuck. “Cancel” (中止) now also takes effect immediately during reading.
1.2.73 (build 113) — November 9, 2026
- When reading subtitles burned into the picture, the app now first test-reads a few places to tell which language the subtitles are written in (English, Korean, Japanese, Chinese and so on) and reads them in that language. In Korean dramas with burned-in English subtitles, misreadings such as “I” becoming “|” or “J”, missing apostrophes (youtre) and symbols turning full-width are reduced.
1.2.72 (build 112) — November 9, 2026
- When translation with Codex stops partway, the notice now shows the reason (the connection dropped, the usage limit was reached, the sign-in expired and so on). For temporary failures such as a dropped connection, the part that stopped is retried once automatically.
1.2.71 (build 111) — November 5, 2026
- The name of the file being worked on is now shown in the progress area and in the window title. While a video is being opened, checked or converted, the name of the file you dropped (or chose) is also shown.
1.2.70 (build 110) — November 4, 2026
- Fixed the whole process stopping with an English error when one image failed to be read while turning image subtitles in a video (Blu-ray or DVD subtitles) into text. A failed image is now read again after a short wait, and only images that still cannot be read are skipped.
1.2.69 (build 109) — November 2, 2026
- “Blur the original subtitles” (元の字幕をぼかす) when exporting a video now blurs only around the times when a new subtitle is shown: from 1 second before a subtitle appears until 0.7 seconds after it disappears. Scenes without subtitles are not blurred. The original subtitles stay unreadable even if you make the subtitle background lighter or turn it off.
1.2.68 (build 108) — November 2, 2026
- When you open a video, the app now also looks for subtitle files in folders below the video’s folder (such as “Subs” or “English”, up to three levels down) and offers them as source candidates. Only files with the same name as the video, or with the same episode number in their name, are used.
1.2.67 (build 107) — November 1, 2026
- Made the subtitle background a consistent height. Even a one-line subtitle now has a background two lines high, with the line shown vertically centred. The same applies to videos exported with burned-in subtitles.
1.2.66 (build 106) — November 1, 2026
- When translating a Korean title from its English subtitles, the Korean audio is now transcribed on this Mac even when translating with Codex. The transcribed Korean is not sent to OpenAI; it is used for the following two things.
- Job titles in the glossary are now corrected to match how people are addressed in the audio (for example, “Director Lee”, which had become イ監督 or イ部長, becomes イ理事 if the audio says “이사”, and “CEO” becomes 代表). Spellings you corrected by hand are not changed.
- Rows where the Korean speech is polite but the translation is casual (タメ口) are now added to “Check with the video” (映像で確認).
1.2.65 (build 105) — November 1, 2026
- When translating a Korean title from its English subtitles with a local AI (LM Studio and so on), the Korean audio is now also transcribed automatically and used as a reference for translating each row. Mix-ups between polite and casual speech are reduced (in a test with 3 episodes, serious errors fell by about 40%). Translation with Codex works as before, and nothing more is sent.
- When a Japanese title is looked up on TMDb, cast members are now shown with their Japanese names and character names written in the Latin alphabet are shown in katakana. They are matched by reading against the character names made from the subtitles, and names whose reading differs slightly are marked “Different reading?” (読みが違う?).
- Source text read from the screen now shows the heading “English subtitles (read from the screen)” (英語の字幕(画面から読み取り)).
1.2.64 (build 104) — October 31, 2026
- After you correct a subtitle in the input field at the bottom, clicking a row in the subtitle list on the right now leaves the input field. Before, the focus stayed in the input field, and pressing N, G or ⌫ typed that character into the subtitle.
1.2.63 (build 103) — October 31, 2026
- After subtitles are read from the screen, rows that look misread are now marked “Check reading” (読み取り要確認) (low text-recognition confidence, garbled duplicates, rows of only numbers or symbols, doubtful English spelling and so on). ⌘↩ confirms a row and ⇧⌘⌫ returns it to before the correction. “Fix with AI” (AIで直す) corrects only the reading errors in one go, and “Remove non-dialogue rows” (台詞でない行を消す) removes on-screen text that is not dialogue in one go.
- The times when subtitles read from the screen appear and disappear are now matched to the original subtitles frame by frame (before, they were up to 0.5 seconds late).
- When you delete a row with ⌫, the video now also moves on to the scene of the next row.
- Instructions for keys such as ⌫, G and ⌘Z are now always shown below “Source / Japanese translation” (原文/日本語訳) in the subtitle list.
1.2.59 (build 99) — October 31, 2026
- Select a subtitle row and press G to extend its display to the end of the next row, keeping the text of the selected row (the next row is removed). When the same subtitle has been read from the screen as several rows, press it repeatedly to merge them into one. ⌘Z undoes it.
- Select a subtitle row and press ⌫ (the delete key) to delete that row and select the next one. Press it repeatedly to delete unwanted rows one after another. ⌘Z undoes it.
1.2.57 (build 97) — October 31, 2026
- “Read subtitles shown on screen” (画面に映っている字幕を読み取る) now lets you set the area to read. Fit the yellow frame shown over the video to where the original subtitles are, and only the text inside the frame is read; signs and captions are not picked up.
- After you set the frame, the app checks the subtitle position in scenes throughout the video and widens the frame if subtitles stick out, or shrinks it if it is too big (“Adjust the frame automatically from the video” (動画を見て枠を自動で直す)). The adjusted frame is used from the start for the next video, so from the second episode of a series you can read straight away.
- When exporting with burned-in subtitles, you can now choose “Blur the original subtitles” (元の字幕をぼかす). Subtitles already burned into the video are blurred across the frame area before the new subtitles are laid over them.
- The subtitle font size can now be made as small as 10 pt (until now, 14 pt).
1.2.55 (build 95) — October 31, 2026
- “Don’t show on screen” (画面に出さない) can now be toggled with the H key. In the Japanese view, select a row and press H to hide it from the screen; press it again to show it again. It can also be chosen from “Review” (確認) in the menu bar.
1.2.54 (build 94) — October 30, 2026
- Fixed rows set to “Don’t show on screen” (画面に出さない) still showing the grey background box on the video even though the subtitle text was gone. The background no longer appears either (if “Always show background” (背景を常に表示) is on in Settings, the band is shown, as in other places without subtitles).
1.2.53 (build 93) — October 30, 2026
- Rows that don’t need translating can now be set to “Don’t show on screen” (画面に出さない). In the Japanese view, select a row and press “Don’t show this row on screen” (この行を画面に出さない) above the editing field; nothing is then shown in Japanese for that row. It is written to the SRT as an empty subtitle and is not shown in exported videos either. It stays that way when you retranslate or review, and you can undo it with “Show on screen” (画面に出す) or by entering a translation.
1.2.52 (build 92) — October 28, 2026
- Fixed a subtitle row consisting only of a quoted title, such as "Fear and Loathing in Las Vegas", coming back in English and causing a translation error. When retranslating, the app now instructs the engine to translate it as the title used when it was released in Japan (or the original title if unknown), enclosed in 『』.
1.2.51 (build 91) — October 28, 2026
- Fixed people appearing as themselves (documentary participants, guests and so on) being missing, or lumped together as “guest”, in titles looked up on TMDb. They are now listed by name as “(themselves)” ((本人)). Titles you looked up on TMDb before are looked up again when opened.
1.2.50 (build 90) — October 28, 2026
- New or small titles that are not on Wikipedia can now also be searched on TMDb (The Movie Database). This needs an API key from your own TMDb account (free for personal, non-commercial use). Choose TMDb under “Search in” (探す先) on the search screen to see how to get a key (how to use). Once a key is saved, automatic lookups also search TMDb when nothing is found on Wikipedia. The key is stored only in this Mac’s keychain, and only the search words, the title’s ID number and the key are sent to TMDb. The TMDb attribution has been added to “About this app” (このアプリについて).
1.2.49 (build 89) — October 28, 2026
- If Movie Metadata (an app by the same author that writes information such as the title and synopsis into videos) is installed, “Open in Movie Metadata” (Movie Metadataで開く) is now shown after a video export finishes. You can open the exported video right away and write the title’s information into it. When exporting in a batch, it is shown for each video.
1.2.48 (build 88) — October 28, 2026
- When you open a video, the app now first asks whether to look up the title on Wikipedia (how to use). Choose “Look it up” (調べる) to open the search screen, where you can correct the title, year and type (film or series) read from the file name and search again. Candidates show a description of the title and the main characters and cast, and if there are subtitles, characters whose names appear in them are marked. The title you choose (or “Don’t look it up” (調べない)) is remembered by title, year and season, so you are not asked again for other episodes of the same series.
- Fixed the app failing to settle on the title’s information for titles in which a character’s short name (such as Maui in Moana) appears many times.
1.2.47 (build 87) — October 27, 2026
- Fixed correct translations being wrongly judged as copies of the example sentences, and discarded every time they were retranslated, when the source contained words such as “greenhouse gases” (温室効果ガス).
1.2.46 (build 86) — October 27, 2026
- Fixed the app failing to settle on the title’s information when several titles were made from the same original work (such as the 2026 film and the 2024 TV film of Sense and Sensibility). It now chooses the title whose release year matches the year in the file name.
1.2.45 (build 85) — October 27, 2026
- Dialogue without subtitles can now be transcribed from the audio and added (“Other” (その他) → “Add unsubtitled dialogue by transcription” (字幕のない会話を文字起こしで足す), how to use). When subtitles leave out scenes where characters speak English, as with the English subtitles of Korean dramas, only those sections are transcribed and added as rows, which can be translated together with the other rows. Sections heard as dialogue in the video’s own language, and sections of music or silence, are not added. Added rows can be compared with the voice under “Rows added from audio” (音声から足した行) in the list.
1.2.44 (build 84) — October 27, 2026
- Fixed looking up a title’s information sometimes being refused every time while Wikipedia’s title data (Wikidata) was busy, leaving the title not found (The Cycle and others).
1.2.43 (build 83) — October 27, 2026
- Fixed how the names of titles looked up on Wikipedia are shown. When an English-language film (such as The Dog Stars) had a Korean title, the Korean was shown first, as in 『도그 스타(The Dog Stars)』; it is now shown with its English title. Korean or Japanese titles are shown first only for titles originally in Korean or Japanese. The display of titles looked up before is also corrected when they are opened.
1.2.42 (build 82) — October 27, 2026
- When a title’s information could not be looked up because Wikipedia was busy or there was no internet connection, the app now automatically looks it up again after about 1, 3 and 10 minutes. The time of the next lookup is shown on the row that shows the title’s information. If all three attempts fail, you can look it up again with “Look it up again” (もう一度調べる), as before.
1.2.41 (build 81) — October 25, 2026
- In the Ollama and LM Studio model lists, Gemma 4 is now also a “★ Recommended” (★ おすすめ) candidate. At the same size, Gemma 4 is preferred (because in the author’s comparison it made fewer serious errors than Qwen). The rule of choosing the largest model that fits in memory first is unchanged.
1.2.40 (build 80) — October 25, 2026
- When translating with Gemma 4 in LM Studio, the “thinking” step before answering is now skipped. In measurements on the author’s Mac, translating 3 rows went from about 29 seconds to about 5 seconds, with no increase in translation errors. Other models, such as Qwen, work as before.
1.2.39 (build 79) — October 23, 2026
- In Korean subtitles, rows whose sentence continues into the next subtitle (such as 「〜しかけた」 or 「〜だったという」) are now retranslated so that the translation continues into the next row, instead of being cut off with “…”. Fixed rows that had been translated correctly on retranslation sometimes being wrongly judged as copies of the example sentences and discarded. The wording of the notice shown when translation ends partway has also been made clearer.
1.2.37 (build 77) — October 21, 2026
- In “Characters” (人物設定) in the work notes, the fields for each character’s first-person pronoun, forms of address, manner of speech and so on are now wider. The whole tab scrolls together, and the input fields are shown over several lines.
1.2.36 (build 76) — October 21, 2026
- The steps at the top of the window are now combined into three: “1 Open Video” (1 動画を開く), “2 Make Japanese subtitles” (2 日本語字幕を作る) and “3 Check and save” (3 確認して保存). Just pressing “2 Make Japanese subtitles” runs transcription (when needed), the work notes and spelling of names, translation and rule-based checks in a row. Transcription only, checking the work notes and so on can be chosen from the arrow to the right of the button.
- “3 Check and save” (3 確認して保存) lists the rows worth checking before you save, if there are any, and otherwise saves the Japanese subtitles. AI rereading, first-person and tone checks, watching through and exporting video can be chosen from the arrow to the right of the button.
- Settings are now split into the tabs “Basic” (基本), “Subtitle appearance” (字幕の見た目) and “Advanced” (詳細). The option to choose the genre of the title has been removed, because choosing it did not change the translation.
- On a Mac using the app for the first time, the engine is chosen automatically: Codex if it can be used, otherwise Ollama or LM Studio with a model that fits. Consent to sending to Codex is now confirmed only once, and no longer asked for each title. With Codex, the approximate number of requests and the time needed for the title are shown.
- Translations of more than 16 characters on one line are now split into two lines at a natural point, such as after a comma or full stop. Lines translated by Codex that are too long for their display time are retranslated shorter, keeping numbers, negations, names and so on.
- Stopping a translation no longer shows an error.
1.2.35 (build 75) — October 20, 2026
- “4 Translate” (4 翻訳) now translates the whole video straight away without setting up characters. For new titles, “Characters” (人物設定) in the work notes is set to “Automatic (no setup needed)” (自動(設定不要)), and manners of speech supported by the subtitles are used only for rows where it is certain who is speaking. If you want to check the title settings and style before translating, as before, you can choose “Use only the checked settings” (確認した設定だけ使う).
- Ollama and LM Studio now also translate 8 rows at a time, reading the surrounding conversation. You can also choose 16 rows, 40 rows or the previous 1 row under “Local translation unit” (ローカル翻訳の単位) in Settings.
- Added “Check meaning and subtitle alignment after generation” (生成後に意味と字幕の対応を確認する) to the Ollama and LM Studio settings (on by default). It checks for mixed-up subjects or negations, wrong digits in amounts, lines mixed in from neighbouring dialogue and so on, and corrects only rows where both the error and the fix are confirmed. It takes longer.
- Added two optional extra analyses to “Characters” (人物設定) in the work notes: “Organise speech by voice on this Mac” (Mac内で声ごとに発言を整理する) (downloads a speaker analysis model from Hugging Face and processes on the Mac) and “Codex image analysis of the video” (Codexで映像の画像を補助解析する) (sends still images taken from the video to OpenAI). Both are off by default.
- Dialogue whose speaker is unknown is no longer given character-like phrasing such as 「俺」, 「僕」 or 「〜わ」.
- Added “Hard to read in time” (読み切りにくい), which collects rows with many characters for their display time, to “Optional checks” (任意の確認) above the list.
- Fixed lines such as 「落ち着いて指示に従え」 and repeated calls such as “Bishop, Bishop.” being treated as translation errors.
1.2.34 (build 67) — October 18, 2026
- Opening the MP4 converted from an MKV is now treated the same as opening the original MKV next to it, and subtitles are read from the MKV. Fixed opening the MP4 keeping the Arabic subtitles chosen by the previous version, and the subtitles in the video not appearing in “Change source…” (原文を替える…).
- Autosaves made while the MP4 was open are carried over as work on the original MKV.
1.2.33 (build 66) — October 18, 2026
- Episodes where Arabic subtitles became the source when opened in 1.2.31 are now switched to more accurate subtitles (English and so on) when reopened. This happens only when the subtitles were chosen by the app itself, no row has been translated yet and no manual edits remain. A notice tells you when they are switched.
- A source you chose yourself, for example with “Change source…” (原文を替える…), is remembered for each video and not switched when it is reopened.
1.2.32 (build 65) — October 18, 2026
- When the subtitles in a video have no language tag, the source is now chosen by judging the language from the subtitle text. Fixed Arabic subtitles sometimes becoming the source even though English subtitles were available.
- Subtitles that have the full dialogue of the whole video are now treated as ordinary subtitles even if they are tagged “forced” (subtitles only for scenes in another language).
- When the only subtitles are in a language different from both the audio and English, they are not loaded; instead the “What should be translated into Japanese?” (何から日本語に訳しますか) screen opens. In a series, the language chosen for an earlier episode is not asked again for the next one.
- “Change source…” (原文を替える…) now lists the subtitles in the video with their language and number of rows, and if subtitles more accurate than the current source have been found, they are preselected. “Export in batch” (まとめて書き出す) and “Read all episodes’ videos to create” (全話の動画を読んで作る) now choose in the same way.
1.2.31 (build 64) — October 17, 2026
- In the check after translation, rows where English (Latin letters) remains in the translation now have just that part corrected to Japanese (such as 「ゴン・EUNテ」 → 「ゴン・ウンテ」). English commonly used in Japanese, such as CEO, SNS and OK, is left as it is.
- In the glossary, spellings where part of a name is still in Latin letters (such as 「ゴン・EUN」) are no longer used in translations, and “Terms & characters” (用語・人物) now notes “Name still contains Latin letters” (名前に英字が残っている).
- The note “About the genre setting” (ジャンル設定について) in the work notes is now shown below the genre field only when the genre is the same as when the work notes were created. Fixed the old note remaining after the genre was changed.
- When the summary of the work notes comes back empty, it is now created again.
1.2.30 (build 63) — October 17, 2026
- Removed “Find wording that doesn’t suit the title” (作品に合わない言葉づかいを探す) from the “5 Review” (5 見直す) menu. In tests it hardly worked: across 160 rows, this Mac’s AI made 1 suggestion and Codex 0. Wording errors are also picked up by “Have the AI reread and suggest” (AIに読み直させて提案を作る).
1.2.29 (build 62) — October 17, 2026
- The confirmation before sending subtitles to Codex is now shown only the first time after opening a video. After that, you are no longer asked when you ask Codex for style samples, the full translation, review and so on for the same video. You are asked again when you open another video or change the model.
1.2.28 (build 61) — October 17, 2026
- The AI reread after translation is now, by default, done only when translating with this Mac’s AI. Codex translations read the flow, so there is little to fix, and in tests more than half of the reread suggestions were worse than the original translation. You can change this with “AI reread after translation” (翻訳後のAIの読み直し) in Settings, and reread at any time from “Review” (見直す).
1.2.27 (build 60) — October 17, 2026
- Removed two experimental items from “How translation proceeds” (翻訳の進め方) in Settings (retranslating sentences that span subtitles together / adding the speaker and the characters in the scene to the translation), because in tests they did not improve the translation. Translating by reading the flow of the subtitles is handled by Codex’s “Translate together” (まとめて訳す).
1.2.26 (build 59) — October 17, 2026
- When Codex is chosen as the translation engine, it now translates 40 rows at a time, reading the surrounding flow. In the film tested, serious mistranslations fell to about one fifth of those from this Mac’s AI.
- Below “Choose translation engine” (翻訳エンジンを選択) in Settings, the differences between this Mac’s AI and Codex (cost, where the subtitle text is sent, translation accuracy, how they translate) are now shown side by side. The screen for checking the style with 5 rows also shows a one-line tip.
1.2.25 (build 58) — October 17, 2026
- While the work notes are being created, “Creating work notes…” (作品メモを作成中…) is now shown in large text just below the buttons, explaining what is being done, that it takes a few minutes, and that a screen for checking them opens when they are ready. The “3 Work notes” (3 作品メモ) button also shows “Creating…” (作成中…) with a spinning mark.
1.2.24 (build 57) — October 17, 2026
- Added “3 Work notes” (3 作品メモ) to the step buttons (translation is now “4 Translate” (4 翻訳) and review “5 Review” (5 見直す)). Before translating, you can check the work notes and spelling of names created by the AI. The conversation style and the genre of the title can also be set here. Finished steps get a ✓, “2 Transcribe” (2 文字起こし), which is not needed when there are subtitles, gets a dotted circle, and the step to press next is underlined.
- Among rows with translation errors, rows that stop short after just a pronoun, such as “She's...” or “I...”, are now translated in a fixed form such as 「彼女は…」 or 「私…」 even if the AI returns only “…” (forms such as 「私」 and 「あなた」 can be checked as “Uncertain translations” (自信のない訳)).
1.2.23 (build 56) — October 17, 2026
- When retranslating rows with translation errors using the subtitles before and after them, rows that break off, such as “She's...” or “I...”, are now translated up to the word they break off at, as in 「彼女は…」 or 「僕は…」. A translation of just “…” is no longer provisionally accepted, and rows that could not be translated stay as translation errors so they are not overlooked.
- Fixed correct translations being rejected as “a number not in the source” for expressions that mean one even though the source has no digit, such as “within a kilometre” (1キロ以内) and “this past month” (この1ヶ月).
1.2.22 (build 55) — October 15, 2026
- Fixed rows in English subtitles beginning with phrases such as “I think” (〜と思う) often being translated as 「てっきり〜と思った」 (the opposite meaning, that one had been mistaken). This is fixed for rows translated from now on.
- Fixed the app sometimes crashing on opening subtitles that include sound effects and speaker names (subtitles for the deaf and hard of hearing) with short rows such as “- I”.
1.2.21 (build 54) — October 15, 2026
- Fixed “Accept all” (すべて採用) and “Reject all” (すべて却下) for suggestions freezing for a long time when there were many (what took 108 seconds with 1,500 rows and 500 suggestions now finishes instantly). Right after you press it, “Accepting N suggestions…” (提案N件を採用しています…) is shown, and when it finishes you are told how many.
1.2.20 (build 53) — October 15, 2026
- When looking up a title on Wikipedia, characters played by actors without an English article, and characters addressed in the subtitles by a title and surname such as “Mr. Yang”, are now counted too. Titles that were not found before (such as The Ordinary Jackpot) can now be found. Titles that were not found before are looked up again. “Look it up again” (もう一度調べる) has been added to the “Not found” (見つかりませんでした) message.
1.2.19 (build 52) — October 15, 2026
- The app now finds English words translated with a meaning that doesn’t fit the surrounding scenes (for example, a row that translated “stock” as 「株」 (shares) when the talk was about a shop’s stock). The check after translation that doesn’t use AI makes a “Word meaning” (語の意味) suggestion, and “Review the translation” (訳文を見直す) always checks it as well. 16 words, including stock, bank, cell and court, are covered.
1.2.18 (build 51) — October 15, 2026
- With “AI used for review” (見直しに使うAI) in the “4 Review” (4 見直す) menu, the review work alone can now be done with a different AI (such as Codex) until you quit the app. Translation still uses the AI in Settings.
1.2.17 (build 50) — October 15, 2026
- After translation, checks that don’t use AI now run in a few seconds: retranslating rows that are not in Japanese, rewriting names that differ slightly from the glossary, marking rows where Korean polite speech was translated as casual speech with “Check with the video” (映像で確認), and so on.
- The AI reread after translation is now, by default, done only when using Codex (you can change this with “AI reread after translation” (翻訳後のAIの読み直し) in Settings).
- For videos where the title can be identified, once you allow it, the app looks up the title on Wikidata and Wikipedia and uses the characters’ names, genders and relationships in the translation. Only the title, year and season are sent.
- Korean names are now written the same way in every row, and names transcribed slightly differently are corrected to the title’s spelling.
- Improved the accuracy of the glossary (honorifics, ranks in Korean titles, plurals, how job titles are used in address; speaker labels, swear words and forms of address are not made into terms).
- Polite sentence endings in North Korean and Gyeongsang dialects are now treated as polite speech.
- Fixed descriptions of sound effects spanning two lines sometimes remaining in the subtitles.
- Fixed the app rarely crashing when you stopped an export with burned-in subtitles.
- As experimental features, added a setting that passes the speaker and the characters in the scene to the translation, and a setting that retranslates sentences spanning several subtitles together (Settings → “How translation proceeds” (翻訳の進め方); off by default).
1.2.16 (build 49) — October 13, 2026
- Fixed exports with burned-in subtitles sometimes stopping partway and never finishing, for example with long 4K or HDR videos. If an export makes no progress at all for 2 minutes or more, it is now stopped and you are told.
1.2.15 (build 48) — October 12, 2026
- When you open an MKV or similar file, the app now shows a list of audio and subtitle tracks before converting, so you can choose the audio to use and how to extract subtitles (don’t extract, source only, source and Japanese translation, Japanese translation only). If you choose Japanese translation only, the Japanese subtitles are loaded as they are and the video export screen opens.
1.2.14 (build 47) — October 11, 2026
- Fixed the video continuing to play when you closed the window during playback. Playback also stops when you close it in viewing mode.
1.2.13 (build 46) — October 11, 2026
- Pressing “3 Translate” (3 翻訳) now first creates the work notes and glossary, lets you check them on the title settings screen, and then goes on to the 5-row style check and the full translation, in that order. The 5-row check also translates using the style policy in the work notes and the glossary.
1.2.12 (build 45) — October 11, 2026
- Pressing “Translate with these settings” (この内容で翻訳) on the title settings screen now goes straight on to the 5-row style check, instead of stopping with the notice “First check 5 rows of translation in the current style with ‘3 Translate’.” (まず「3 翻訳」で、現在の文体の訳を5行確認してください。).
1.2.11 (build 44) — October 11, 2026
- Added viewing mode. “Watch” (鑑賞する) (⇧⌘F) shows the video filling the window, so you can watch it through with subtitles the same size as in the exported video. If a subtitle looks wrong, press Return to go back and fix it.
- Added “Keep the background shown where there are no subtitles” (字幕のないところでも背景を出したままにする) to “Subtitle appearance” (字幕の見た目) in Settings. The background stays on screen as a band two lines high, and the subtitles change on top of it. This is also included in videos exported with burned-in subtitles.
- The subtitle position and the fine vertical adjustment are now kept the next time you open the app.
1.2.10 (build 43) — October 7, 2026
- Fixed translations imported from a film’s Japanese subtitles and saved not being loaded in full when the video was reopened. This happened when there were rows with line breaks between speakers without “-”, or rows with converted units such as 「25メートル」 (25 metres). Japanese subtitles saved by this app are now loaded exactly as saved if the number of rows and the times match the source.
- When importing another translated SRT with “Load translated SRT…” (訳文SRTを読み込む…), the speaker breaks and the content of the translation are checked as before.
- In the “Export in batch” (まとめて書き出す) list, videos whose source subtitles cannot be found now show the reason more clearly.
1.2.9 (build 42) — October 7, 2026
- Fixed the Japanese translation not being loaded at all when a video was reopened, with only the source shown, if the saved Japanese subtitles had rows with short translations of two people’s short lines (such as 「- え?」 and 「- 今?」). The length of the translation is now checked for each speaker.
- If saved Japanese subtitles cannot be matched with the source and loaded, the file and the reason are now shown when you open the video (the file is left as it is).
1.2.8 (build 41) — October 7, 2026
- Added “Export in batch…” (まとめて書き出す…), which turns every episode of a series, or several films, into subtitled videos one by one with the same settings (“File” (ファイル) menu, ⇧⌘E). Episodes of the same series in the same folder as the open episode are listed in episode order, and for each video it shows whether it is ready, the number of rows without a translation, unsaved changes and whether it has already been exported. You can keep translating and checking other videos during the export.
- When exporting in a batch, if one video cannot be exported the app moves on to the next, and shows all the results at the end. Even if you stop partway, already exported videos are left out when you open it again, so you can continue from where you left off. MKV files without a converted copy are converted before exporting, and the converted copy is deleted after exporting.
- The Mac no longer goes to sleep while a video is being exported.
1.2.7 (build 40) — October 6, 2026
- Work notes and glossaries can now be shared across episodes of a series. If the name of a video or subtitle file has an episode marker (S01E02, 1x02, 第2話 and so on), one set of work notes and one glossary are used for all episodes of the title. When you open a new episode, the spellings of names and terms and the characters’ manners of speech decided in earlier episodes are carried over, and only words appearing for the first time in that episode are newly decided. Spelling corrections made in any episode are used in the other episodes too.
- Added “Read all episodes’ videos to create…” (全話の動画を読んで作る…) (choose the videos or subtitle files of all episodes at once and create the work notes and glossary first), “Read this episode and add” (この回を読んで足す) (adds new characters and the synopsis of that episode) and a “Share with other episodes” (ほかの回と共有する) checkbox to the title settings screen.
- On-screen text in capitals only (such as “TOTAL WINNINGS”) is no longer picked up as glossary terms.
1.2.6 (build 39) — October 6, 2026
- Added “Check tone row by row” (口調を1行ずつ確かめる) to “4 Review” (4 見直す). Dialogue rows are labelled with the tone determined from the sentence ending (polite です・ます / casual タメ口). For rows that don’t fit, such as a subordinate speaking casually to a boss, ⌘5 (です・ます) and ⌘6 (タメ口) create a version with only the sentence ending rewritten, which you can accept or reject. ⌥⌘5 / ⌥⌘6 make a whole scene consistent.
- Fixed being unable to go further if any row in the trial translation of the first 5 rows in “3 Translate” (3 翻訳) was not accepted. The trial translation now skips title cards and on-screen text and picks rows from the dialogue, and translations that copy the source’s numbers as they are are no longer judged too long.
1.2.5 (build 38) — October 6, 2026
- Introduced Terms of Use. ことばシンク is a tool for adding Japanese subtitles to videos such as your own recordings or recordings of international conferences and talks, to watch them yourself. Publishing, distributing or selling subtitles or videos made for other people’s copyrighted works without permission, and commercial use, are prohibited.
- A Terms of Use screen appears at launch; please agree to it (the app cannot be used if you do not agree). The Terms of Use can be read at any time with “Terms of Use…” (利用規約…) in the app menu.
1.2.4 (build 37) — October 6, 2026
- When you select a row with a suggestion, the suggested text is now shown in the video preview on the left, looking just as it would if accepted. A “Suggestion preview (not accepted)” (提案のプレビュー(未採用)) label appears at the top left. During normal playback, the current translation is shown.
- Fixed the cursor jumping to the end when you corrected text in the middle of a sentence in the editing field. The same fix was made to the “Manner of speech” (話し方) field in the glossary.
1.2.3 (build 36) — October 6, 2026
- “Next step” (次にすること) is now a large card just below the buttons. It shows the next task alongside what happens if you don’t do it (such as “You may save without fixing them. The 3 rows without a translation will appear in English in the subtitles” (直さずに保存しても構いません。訳のない 3 行は英語のまま字幕に出ます) and “Save as is…” (このまま保存…)). When the tasks before saving are done, saving becomes the main button.
- Videos with burned-in subtitles are now exported at about the same size as the original video (a 2-hour film that was 7.6 GB is now 2.0 GB). They are exported in HEVC (H.265), with the audio of the original video included as it is. On the export screen you can choose “About the same size as the original” (元の動画と同じくらいの大きさ), “High quality” (高画質) or “Compatibility first (H.264)” (互換性重視(H.264)). Exporting is also faster (about 5–10 minutes for a 2-hour film).
- Added “Check first-person pronouns row by row” (一人称を1行ずつ確かめる) to “4 Review” (4 見直す). When, for example, a woman’s line uses 「僕」, you can fix it while listening to the voice with ⌘1 私, ⌘2 あたし, ⌘3 僕 and ⌘4 俺.
- When checking suggestions, the list now shows the “Current translation” (今の訳) and the “Suggestion” (提案) one above the other. When you move to the next row with a key, the list scrolls along.
- The notice that translation is complete now shows the number of uncertain translations. Uncertain translations are also included in the check before saving.
- Fixed saving only the source subtitles showing that the Japanese translation had been saved too. Fixed the last frame of a video with burned-in subtitles sometimes staying frozen for several seconds.
1.2.2 (build 35) — October 5, 2026
- The counts above the list are now split into two rows in the order of work. “Before saving” (保存前に) (translation errors, uncertain translations, “Check what was heard” (聞き取り要確認), untranslated) are rows that would put English or mistranslations in the subtitles if left; “Optional checks” (任意の確認) (import mismatches, suggestions, rows the AI fixed) are rows you can save without looking at all of them. Items with 0 are not shown.
- Click a count to see, above the list, what those rows are, what to do with them, what happens if you leave them and how many are done. The “?” to the right of the counts and “How to read the counts and marks in the list” (一覧の件数と印の見方) in the Help menu explain all the counts and marks. The marks on rows now have the same shape and colour as the counts.
- Made names easier to understand (“Tidied rows” (整えた行) → “Rows the AI fixed” (AIが直した行), “Source check on hold” (原文確認の保留) → “Check what was heard” (聞き取り要確認), “Translation not entered” (訳文未入力) → “Untranslated” (未翻訳), and “Needs checking” (要確認) for imports → “Import mismatches” (取り込みのずれ)). Rows that could not meet the conditions however many times they were retranslated are counted as “Uncertain translations” (自信のない訳). Accepted suggestions are not counted as awaiting review.
- Added “Review” (確認) to the menu bar. With ⌘↩ “Confirm and next” (確認して次へ), ⇧⌘⌫ “Reject / revert and next” (却下・元に戻して次へ) and ⌘] / ⌘[ “Next / previous unhandled” (次・前の未処理へ), you can clear the counted rows one by one with keys. R plays just that row, and ⇧⌘R plays the unhandled rows in a row (check playback).
- Added “Mark all as checked” (まとめて確認済みにする) for import mismatches and rows the AI fixed, and “Merge with the neighbouring row without a translation” (訳なしの隣の行とまとめる) for Japanese subtitles spanning two English rows (can be undone right afterwards).
- “Next step” (次にすること) now looks at the counts in the order of work to show the next task. When subtitles have been imported, it suggests checking import mismatches before translating. Imported Japanese is not rewritten by later reviews or spelling unification.
- When saving the Japanese translation or exporting a video, you are now told that rows without a translation will appear in English (the source).
- Style, genre and notes on manner of speech are now remembered for each video. When you open another video, the work on the previous title (filters, search, translation in progress, playback speed and so on) returns to its initial state.
1.2.1 (build 34) — October 5, 2026
- The app now updates itself. It checks for a new version once a day, and if it finds one it downloads it in the background and checks the SHA-256, the Developer ID signature and Apple’s notarization. It never restarts while you are working; it is replaced the next time you quit the app. The next time you open it, you are told that it was updated and what changed.
- To update right away, choose “Restart and Update Now” (今すぐ再起動して更新) from “Check for Updates…” (アップデートを確認…) in the app menu. If you have unsaved subtitles, it checks with you first.
- Added “Install Updates Automatically” (アップデートを自動でインストール) to the app menu. If you turn it off, you only get a notice of new versions, as before.
1.2.0 (build 33) — October 5, 2026
- First public release. You can transcribe subtitles from a video, correct the text and times while it plays, add a Japanese translation and save it as SRT.
- Added “About this app” (このアプリについて). From the app menu you can see the version, the copyright notice (© 2026 K386) and a link to the official site. It is shown once on first launch and once on the first launch after updating to a new version.
- Added the check for new versions. Once a day it asks the official site and tells you if there is a new version. Press “Update and Relaunch” (更新して再起動) to have it check the signature and notarization and then replace the app. It can be turned off with “Check for Updates Automatically” (アップデートを自動で確認) in the app menu.
- Added the app icon.