AI utilities — context-aware subtitle translation

Tools that process just the files you upload, right away, without creating a new project. The only one available now is context-aware subtitle translation; the rest are shown on screen as Coming soon.

Not a line-by-line translation

Ordinary subtitle translation goes line by line, so the same character switches between formal and casual speech and names are spelled differently from file to file.

Here the whole series is read first, the rules are set, and then it translates. There are no questions to answer — it goes ahead with what the AI inferred.

1. Series rules — reads every subtitle and sets the synopsis, character relationships, formality and the spelling of proper names for the whole series.

How characters address each other is also fixed to a single target-language form used in every episode (for example, 아버지 → Dad). A character's gender is set only when the dialogue gives evidence for it; otherwise the translation avoids gendered words such as he, she or Sir and uses the name or role (Teacher) instead.

1. Surrounding context — each file is translated with the flow of the conversation around every line. It first infers who is

speaking to whom, so formality matches the relationship. Across episodes it carries over the story so far, changes in relationships and forms of address, and the name and term spellings chosen in earlier episodes.

1. Subtitle length — a line over the per-line character limit is first split into two lines at a natural break; lines still

too long to read on screen (line count, reading speed) are shortened while keeping the meaning. Timecodes are never changed.

Supported languages

Subtitle translation (series analysis, per-episode translation, shortening long lines) runs on Gemini 3.8 Flash.

Korean · English · Chinese (Simplified) · Chinese (Traditional, Taiwan usage) · Japanese · French · Spanish · Portuguese (Brazil) · German · Indonesian · Thai · Vietnamese · Hindi (13 languages). One source can be translated into several languages at once (language packs). The series is analyzed once, with the first language you pick; the other languages only get their own spellings of names and terms. Each language is translated in parallel. Length limits differ by language (for example 42 characters per line in English, 16 in Korean, 13 in Japanese). The limits for Portuguese, German, Indonesian, Thai, Vietnamese and Hindi are still estimates and will be tuned from real results.

You can run up to 4 translations at once. The languages you pick together count as one, however many there are. If 4 are already running, a new translation stops before the analysis starts and no credits are deducted.

How to use it

It's a single page — upload at the top, download from the "Translation jobs" list below.

1. Upload — drop your subtitle files (SRT · VTT, up to 100 files, 500 KB per file, 5 MB total). Episode order and the series

title come from the file names. Files that aren't UTF-8 (such as EUC-KR from Korean Windows) are converted automatically.

1. Pick languages — the source language is pre-selected from the subtitles (change it if it's wrong). Pick one or more target languages.

• Reference material (optional) — attach a pitch deck, script or notes with "Add material" and they're read alongside the subtitles

during the series analysis, pinning down ages, relationships, forms of address, name spellings and proper nouns. PDF, Word (.docx) or text (.txt), up to 3 files of 10 MB each; about 150,000 characters in total are read, and only the beginning beyond that. Scanned (image) PDFs can't be read yet. The material is used only for the analysis and not stored, and where it differs from the subtitles, the subtitles win. No extra cost.

1. "Start translating" — the button shows the total credits. It reads all the subtitles, sets the rules (1–3 minutes) and then

queues the translation. Uploading files alone doesn't start anything.

1. Download — the "Translation jobs" list shows one card per series and one row per language with its progress. Download a

finished language with its button, or several with "Download all". Translation keeps going on the server even if you close the page. Progress and completion also appear in the generation indicator at the top; "View subtitles" brings you back to this page with that job highlighted.

There's no on-screen editing of the results. Edit the downloaded files in your subtitle editor.

Cost

Example: translating one source into two languages costs analysis 100 + language pack 5 + translation (10 per 1,000 characters × 2 languages). Failed jobs aren't charged. See Credits and generation settings for details.

Good to know

• It's separate from your projects. Files processed here aren't saved to any project or episode —

it's a tool for processing just the files you need and taking them with you.

• If an episode fails with a temporary error, that episode alone is retried up to two more times, then once more with the

subtitles split into smaller pieces (always the same Gemini model). If it still fails, only that episode is left empty and the following episodes keep translating — they carry over the story, forms of address and terms from the nearest earlier finished episode. Credits for the failed episode are refunded, and "Retry failed episodes" in the list translates just that episode again.

• "Start translating" first compares the total on the button with your balance. If it's short, nothing is deducted and you're shown the

credits needed and available.

• If only some languages fail to start, you're told which ones and the rest carry on. Those languages stay in the list with the reason

(for example "not enough credits"), and after topping up, "Start again" starts just those languages (the analysis and language packs aren't redone).