To translate an SRT with AI, open a copy of the subtitles in Subtitle Edit, choose a translation engine in Auto-translate, set the source and target languages, and apply the translation after reviewing it. Save the result as a separate SRT. If your goal is to change only the language, the block numbers and display times must stay the same; also check the meaning and readability against the video separately.
This guide is intended for a marketer or training specialist who already has a video and a synchronized SRT. It does not create subtitles from audio and does not fix subtitles that were mis-synced from the start. The workflow described here was checked against the documentation on October 31, 2026. Translation through the service and the interface behavior were not tested here; the sections below were prepared manually.
Contents: Step 1. Open a copy and review the source subtitles · Step 3. Translate and apply the result · Step 5. Save a separate SRT and compare the structure
What you need and the result to aim for
Prepare the source video, its SRT, installed Subtitle Edit, and an available translation engine. A paid API may require a separate account and key. Installing the editor does not give you a subscription to third-party models.
The result should be two files: the untouched original and the translated SRT, which opens with the same video. For example, lesson.en.srt and lesson.ru.srt. You need to determine translation rights for the video and whether it is allowed to send the dialogue to an external service before uploading.
SRT is a text format. A block contains a sequential number, a display interval, text, and a blank line. Time is written as hours, minutes, seconds, and milliseconds; the timecode uses a comma. The structure is described by the Library of Congress.
Sample source:
1
00:00:01,000 --> 00:00:03,500
Open the customer card.
2
00:00:04,000 --> 00:00:06,000
Select the delivery date.Step 1. Open a copy and review the source subtitles
Create a working copy of the SRT. Open it in the editor and load the matching video. Check the beginning, middle, and end: does each selected line appear where it is spoken? If the text is already lagging behind the speech, fix the source sync first. Translation should not hide this problem.
Look at the block numbers and the number of entries. Note the original count so you can spot anything missing later. In the example above, there are two blocks. In a real file, the number will be different — do not use the training value as a benchmark.
Check Cyrillic text, special characters, names, and the source language. Incorrectly decoded text such as garbled characters must be fixed before sending it to the model: the translator will receive exactly the string that opens.
Step 2. Choose the translation method and where to process it
The current Subtitle Edit documentation lists cloud and local engines. For example, DeepL requires an API key, while a local option is available through Ollama. Their settings differ; seeing a name in the menu does not prove that the model is already downloaded and running on your computer.
| Task condition | What to choose for the first pass | What to confirm before launching |
|---|---|---|
| The dialogue may be shared with an external service | Available cloud engine | Account, key, processing terms, and payment terms |
| The text must be processed in its own environment | Suitable local engine | Installation, model, connection address, and resources |
| You need to fix several lines | Translation of selected lines | Which lines are selected and where the result will go |
These are criteria for choosing a working path, not a ranking of translation quality. For employee instructions, it is more important to translate your system term correctly than to get a polished sentence with a different action. Prepare a small list of approved names: “customer card,” “delivery date,” “order.” Do not assume that any engine will automatically get the corporate glossary.
Step 3. Translate and apply the result
Open Translate → Auto-translate.... Select the engine, the source language, and the target language, then click Translate. Review the result in the table and click OK, to apply it. This is the exact sequence indicated by the Auto Translate guide. The names and placement of the controls may differ in your installed version and localization.
Before working with a large file, it helps to test a small continuous segment with a term, a name, and several neighboring lines. For the selected lines, the documentation describes the Auto-translate command in the context menu. Make sure you translated the right range and did not mistake a partial result for the whole file.
If a local engine was selected and it does not respond, check whether it is running and review the connection settings. Do not quietly switch to the cloud if only local processing was agreed for this text.
Step 4. Adjust the meaning using neighboring lines
Read the sentences as complete units. One subtitle line may begin in one block and end in the next. If you edit each line without context, you can lose a negation or the connection between a pronoun and the object it refers to.
A sample translation of two blocks looks like this:
1
00:00:01,000 --> 00:00:03,500
Open the customer card.
2
00:00:04,000 --> 00:00:06,000
Select the delivery date.Here the text has changed, but the intervals and numbers have been preserved. This is a manually prepared structure sample, not the result of an AI prompt. For your own material, separately verify numbers, button names, prohibitions, names, and units of measure. When translating “do not delete,” missing the negation changes the employee’s action — polished Russian text does not make up for that mistake.
If the automatic translation is longer than the source text, first shorten the wording without losing meaning. Do not automatically extend the block time: it may overlap with the next line. If you cannot avoid changing the time, that becomes a separate sync edit that must be checked on video.
Step 5. Save a separate SRT and compare the structure
Save the result under a new name in SubRip/SRT format with an encoding your player reads correctly; for Cyrillic, UTF-8 is usually chosen. Then reopen the saved file. This helps you catch an incorrect format or encoding while the original is still available.
For a “translation only” task, compare the source and the result by each block number. The number of entries, the order of the numbers, and both time stamps must match. Within the text portion, the number of lines may change because of wrapping; this is different from deleting an entire block.
Do not compare files only by byte size: Russian text and a different encoding produce a different length without breaking the timing. No error on opening is not enough either: a syntactically valid SRT can still contain a missing line.
Step 6. Review the subtitles on the final video
Attach the new SRT to the same video. Check the beginning, middle, and end, then spots with fast dialogue, long terms, and speaker changes. A successful result means the line matches the speech, can be read in time, and is not cut off by the player you chose.
There is no universal character limit set here: requirements can differ across the learning platform, video service, and internal player. Follow the publishing venue’s rules and the actual viewing experience. If the source subtitles are available for users with hearing impairments, preserve meaningful sound and speaker cues instead of removing them for the sake of brevity.
Common mistakes and how to fix them
| Symptom | Possible cause | Action |
|---|---|---|
| Entries disappeared after translation | Only the selected range was processed, or the result was applied incorrectly | Compare the numbering with the original and restore the missing blocks |
| Russian letters are corrupted | Incorrect encoding | Open a copy with the correct encoding and save it again |
| The translation appeared too late | The source was already out of sync or the timecodes were changed | Compare the intervals with the original and check the video |
| The term is translated differently | Segments were translated without consistent terminology | Find the variations throughout the file and standardize the chosen term |
| The line cannot be read within the display time | The translation became longer | Shorten the text or review the timing separately |
This table contains diagnostic hypotheses, not a log of actual detected errors. General approaches to corporate translation are covered in the article AI Translator for Business.
How Vladimir Lomtev’s AI Studio helps with video translation
Vladimir Lomtev’s AI Studio can help organize subtitle work in corporate content:
- prepare source materials and approved naming for recurring training videos;
- connect the translation to selected corporate AI tools;
- set up result review and train employees on the workflow.
To get started, choose one video with the original SRT and specify where it will be published and whether the text can be sent to an external service. Discuss the project.
Bottom line: a new language without accidental timing changes
Work from a copy, translate it through the chosen engine, correct the meaning using nearby lines, and save a separate SRT. For a language swap, the numbers and timecodes must match the original. Finish by reviewing the video: this reveals readability issues that file comparison does not show.
