Free SRT Translator

An SRT translator changes the language of the text in a subtitle file while leaving the timestamps alone, so the new captions still line up with the video. This one translates SRT or VTT subtitles into English or Simplified Chinese and gives you an SRT file back.

1. Click Choose files and open your .srt or .vtt file, or paste the cues. 2. Set “Translate to” to Chinese or English. 3. Click Translate subtitles and wait for every batch to finish. 4. Review the text, then click Download SRT.

Yes. The tool reads each cue’s start and end time, translates only the text and writes the same times into the new file. If the original file was out of sync with your video, the translated one will be too, so fix timing first.

You can translate into English or Simplified Chinese. It’s built for English-to-Chinese and Chinese-to-English subtitles, and other target languages aren’t offered.

Each run handles up to 100 cues, up to 1,200 characters per cue and 16,000 characters in total. For a longer video, split the SRT into parts, translate each part and join them again in your video editor or subtitle app.

Yes. You can open .vtt files or paste WebVTT text, and the header, NOTE and STYLE blocks are skipped. The translated result is always saved as an SRT file, which most video editors and YouTube accept.

No. The translation model runs in your browser tab, so the subtitle text stays on your device. Only the model is downloaded, once, the first time you translate.

It’s free and doesn’t require signup or login. There’s no daily quota built into the tool; the limits are 100 cues per run and what your device can handle.

Yes, in a browser with WebGPU. Phones and lower-memory devices automatically use the lighter Qwen3-1.7B (first download about 1 GB), while computers with enough graphics memory use Qwen3-4B (about 2.3 GB). The lighter model is less nuanced, so check names and wording more closely. It runs fastest in the latest Chrome or Edge on a computer, and if WebGPU isn’t available, the tool suggests opening the page there.

The model reads cues in small groups for context and is told to keep meaning, names and numbers, so most lines read naturally. It can still mistranslate slang, product names or technical terms. Review every cue, especially names, numbers and dates, before you publish.