Translate a whole season of subtitles at once

Translate a whole season of subtitles at once

November 15, 2026
Summary
  • Put every episode's subtitle file in one ZIP and translate the archive once per target language: a subtitle file holds one language, an archive holds up to 200 files.
  • Write the character and place names into a glossary before the first run, because a name that appears in one cue has nothing around it to say it is a name.
  • Run one language first and check it, then loop the rest with the aiglot CLI or hand the loop to an AI agent; only the approve step spends credits.

Put every episode in one ZIP, and translate that ZIP once per language. Ten episodes into five languages is five jobs, not fifty. Each job takes the whole season, applies one instruction and one glossary to every file, and gives you back a ZIP with the same file names and the same timecodes.

That is the whole method. The rest of this guide is the part that decides whether it goes well: how to prepare the folder, which names to write down before you start, how to check the first language, and how to run the others with a short loop or a prompt to your AI agent.

Why languages are jobs and episodes are not

Two facts about subtitle files shape everything.

A subtitle file holds one language. An SRT is a list of cues, each with a number, a timecode and some text. There is no second field for Spanish. So translating it replaces the text in place, and a second language is a second file.

A ZIP holds many files of the same kind. AI Glot unpacks it, translates every file with the same rules, and packs it again with the same folders and names. Up to 200 files and 20 MB per archive, which is a lot of seasons: subtitles are plain text, and an episode is usually tens of kilobytes.

Put the two together and the count of jobs is the count of languages. The count of episodes stops mattering.

Episodes travel together in one archive, so the number of jobs is the number of languages.

Step 1: prepare the season folder

Export every episode’s subtitles from your editor or platform, then put them in one folder. A few rules make the rest painless:

  • One format per archive. All .srt, or all .vtt, or all .ass. A mixed archive is refused, because the whole job is planned as one shape. If your editor gave you a mix, make one archive per format.
  • English files only. Every language starts from the original, never from another translation. A translation of a translation carries the first one’s compromises forward.
  • Name files so they sort. s01e01.srt, s01e02.srt. The names come back unchanged, so they are worth choosing now.
  • Each file up to 4 MB, the archive up to 20 MB, no ZIP inside the ZIP.

Then zip it. On a Mac or Linux:

cd ~/Videos/my-show
zip -r season-1.zip season-1 -x '*.DS_Store'

System files like .DS_Store are left out on our side anyway, but there is no reason to send them.

Step 2: write the names down before the first run

This is the step people skip, and it is the one that shows in episode 7.

A name that appears once has nothing around it to say it is a name. The engine translates each cue with your instructions and the cues around it in view. That is plenty for normal dialogue. It is not enough for a place called “The Hollow” that a character mentions once in episode 3. Nothing in that cue says it is a proper noun, so it can come back as a French common noun in one episode and stay in English in another.

A name that appears once has no context. The glossary decides it once for every episode.

A glossary fixes that. It belongs to one language pair, it is applied to every file in the archive, and it stays in your workspace for the next season. Write a two-column CSV:

source,target
The Hollow,The Hollow
Captain Reyes,capitaine Reyes
Marrow Line,ligne Marrow

Then create one glossary per target language:

aiglot glossaries create en-fr --from names-fr.csv
aiglot glossaries create en-de --from names-de.csv

Two practical notes. The pair has to match the plan exactly: if the plan reports the source as en-US rather than en, name the glossary en-US-fr. And plans differ in how many glossaries you can keep: Free has one, Starter three, Pro unlimited. For five languages, that matters.

Step 3: write two short instructions

There are two places to say what you want, and each one only works for its own kind of rule.

The plan instruction says what to translate. It is read once, against the whole archive. The target language and the scope go here:

The source language is English. Translate the text of every cue in all
the subtitle files into French. Leave cue numbers and timecodes as they are.

The line instruction says how to write each line. It is applied while each cue is written, so it can only describe what is visible in one cue:

Use a natural, spoken tone. Keep each line under 42 characters and use at
most two lines per cue. Prefer a shorter phrasing when a line would not fit.

If you swap them, nothing errors. A line-length rule in the plan instruction is not there when the cue is written, so it does nothing. Keep the scope in the first and the style in the second.

The 42 characters is Netflix’s English guideline for one line. It is a request that shapes the wording, not a check that rejects a long line, so the first file is worth opening.

Step 4: run one language and check it

Start with the language you can read best. The CLI does it in three commands:

aiglot batches create season-1.zip \
  --instruction "The source language is English. Translate the text of every cue in all the subtitle files into French. Leave cue numbers and timecodes as they are." \
  --json | jq '{id, plan}'

That call is free. It reads the archive and returns the plan: what will be translated, what is left alone, the word count and the cost. Read it. If it is wrong, refine it for free with aiglot batches plan <id> --refine "...".

When the plan is right:

aiglot batches approve <id> --quality standard \
  --instructions "Use a natural, spoken tone. Keep each line under 42 characters and use at most two lines per cue."
aiglot batches get <id>        # repeat until the status is completed
aiglot batches download <id> --output season-1.fr.zip

approve is the only command that spends credits, and it asks for confirmation first.

One language, end to end. Everything before approval is free, so the plan can be read and corrected first.

Then check one episode before you run the others. Three checks find almost everything:

  1. Count the cues. The original and the translation of one episode should have the same number of cues and the same last timecode.
  2. Watch a fast minute. Load the track on the video and pick a minute where people talk quickly. Long lines show there first.
  3. Search for two names. One from the glossary and one that is not in it. If the second one drifted, add it to the glossary now, before the other languages.

The full checklist is in the guide to translating subtitles into another language. Everything you fix here, you fix once instead of five times.

Step 5: run the other languages

Once one language is right, the others are the same command with a different word in it. This loop runs them one after another and stops if a job fails:

# Wait until a translation has finished, or stop if it failed
wait_done() {
  while true; do
    status=$(aiglot batches get "$1" --json | jq -r '.status')
    case "$status" in
      completed) return 0 ;;
      failed|cancelled) echo "$1 ended as $status" >&2; return 1 ;;
    esac
    sleep 10
  done
}

for pair in de:German es:Spanish it:Italian pt:Portuguese; do
  code=${pair%%:*}
  lang=${pair#*:}

  id=$(aiglot batches create season-1.zip \
        --idempotency-key "season-1-${code}" \
        --instruction "The source language is English. Translate the text of every cue in all the subtitle files into ${lang}. Leave cue numbers and timecodes as they are." \
        --json | jq -r '.id')

  aiglot batches approve "$id" --quality standard --yes \
    --instructions "Use a natural, spoken tone. Keep each line under 42 characters and use at most two lines per cue."

  wait_done "$id" && aiglot batches download "$id" --output "season-1.${code}.zip"
done

Two details in it are worth keeping, even in your own scripts.

--idempotency-key makes a re-run safe. If your laptop sleeps halfway through, run the loop again. A language that was already created comes back as the same translation instead of a second one.

wait_done also stops on failed and cancelled. A loop that waits for completed alone waits forever when a job fails, while this one stops and tells you.

The --yes skips the confirmation, so only use it once you have read a plan for the same archive. If you want to see every plan before anything is spent, drop the approve line and run it per language after reading.

One English archive, five translated archives. File names and timecodes stay the same inside each one.

Or ask your AI agent to do it

If you use Claude Code, Codex or Cursor in a terminal, you don’t need to write the loop. Give the agent the job in plain words, and it runs the same commands:

My English subtitles for season 1 are in ~/Videos/my-show/season-1.zip,
one SRT per episode.

Using the aiglot CLI, translate the archive into German, Spanish, Italian
and Portuguese: one batch per language, because a subtitle file holds one
language. Read `aiglot batches create --help` first so you use the real
flags. Say "The source language is English" in every instruction.

Line instruction at approval: natural spoken tone, under 42 characters
per line, two lines per cue at most.

Show me the four plans with their word counts and costs, then stop and
wait for my go before approving anything. After that, download each
result as season-1.<code>.zip and tell me if any job failed.

The important line is the stop before approval. Everything up to it is free, so the agent can create, read and re-plan as often as it needs, and you decide on the spend with four costs in front of you. The guide to YouTube subtitles in several languages explains each line of a prompt like this one in more detail.

What a season costs

The cost is the word count of the season, times the number of languages.

Standard Lite
Credits 1 per word 1 per 3 words
Who it is for Dialogue people will watch closely A first pass, or content you will review anyway
Plan needed Any, including the free sample A paid plan

So if your season is 30,000 words, five languages is 150,000 credits in Standard, or 50,000 in Lite. You don’t need to estimate it yourself: each plan shows the measured word count and cost for your own archive before you approve.

Your first translation includes a free sample of 5,000 words, which is enough to see what one episode looks like in another language. The ZIP translator also takes an archive up to 10 MB without an account, five files a day.

Putting the files back

The files come back with their original names inside each language’s ZIP. Most platforms want the language in the file name, so add it before you upload:

mkdir fr && unzip -q season-1.fr.zip -d fr
cd fr/season-1
for f in *.srt; do mv "$f" "${f%.srt}.fr.srt"; done

Then upload each file as a new language on its episode. Every player and platform works one language at a time, so this part does not get shorter, but with the files named s01e04.fr.srt it is at least hard to get wrong.

Where a person still belongs

For a short film in one language, ask someone who speaks it. They’ll do better than any process, and it will take them an afternoon.

For a season in five languages, the file handling is most of the work, and it is not language work. Run the season through AI Glot, then spend the expert time where viewers notice: the first episode, the jokes, and any scene where the wording is the story. Start with one language on the subtitle translator, or create a free account and send the whole season.


Limits quoted here were checked on 15 November 2026: 200 files and 20 MB per archive, 4 MB per subtitle file, and one target language per job. The 30,000-word season is an example for the arithmetic, not a measured file. Your own plan always shows the real figures.

Ready to translate your large files?

Create a free account →

Prompt copied

  1. 1Open your AI agent
  2. 2Paste the promptCtrl V
  3. 3Follow its guidance
Show the prompt

Translate a file for me with AI Glot. Ask me which file and which languages, then follow https://ai-glot.com/agent.md

Your first 5,000 words free