How to Translate SRT Subtitles (and Keep Them Readable)

An SRT is a plain text file made of numbered blocks called cues. Each cue has three parts: a sequence number, a timecode line like 00:01:12,400 --> 00:01:15,050, and one or two lines of text. A blank line separates one cue from the next.

Only the text lines get translated. The number and the timecode line must stay exactly as they are, including the comma before the milliseconds and the arrow. If you paste a whole SRT into a general translator, it can translate or reformat those lines too, and the file stops loading or drifts out of sync. That is the most common failure the ranking guides warn about, and they are right.

This workflow works with any tool, free or paid. The order matters more than the tool.

Sync is not the same as readability. The timecode tells the player when to show a line. It does not care whether a viewer can read it in that time.

Netflix's timed text guides are a good public benchmark. For English they allow 42 characters per line, two lines max, and up to 20 characters per second for adult content. For Spanish (Latin America and Spain) and Brazilian Portuguese the line limit is the same 42 characters, but reading speed drops to 17 characters per second. Their general rules also cap any single subtitle at 7 seconds and keep it on screen at least five sixths of a second.

Here is why that matters. English to Spanish, German or French usually comes out longer. A 2 second English cue with 34 characters sits at 17 cps, which is fine. Translate it to a 44 character Spanish line and you are at 22 cps on a 17 cps budget. The timing did not change, but now the viewer misses half of it.

For short vertical clips with burned-in captions the budget is even tighter, since captions often sit in a narrow column. If you plan to burn captions into Shorts or Reels, check where they land with our guide to safe zones for Shorts, TikTok and Reels.

Speech is cut into cues by time, not by grammar. One English sentence is often split across two or three cues, and word order changes between languages. If you translate each cue alone, you get broken grammar: a verb that belongs at the end of the sentence lands in the first cue, or an adjective ends up stranded.

The fix: read the whole sentence across its cues, translate it as one thought, then redistribute the words so each cue ends at a natural break (after a comma, before a conjunction). Keep the cue count and timecodes, only move words between neighboring cues. Good subtitle editors show the original next to the translation, which makes this fast.

When a line will not fit, try these in order. Most fixes are the first two.

Machine translation (DeepL, Google, or an LLM) is fine for a first draft. It handles common phrases well and saves hours. What it does not do on its own: keep names and brand terms consistent, catch jokes or slang, pick formal vs informal address, or know that a line runs 6 characters over its time budget.

Our take: let the machine draft, then have someone who speaks the target language watch the video with the track on. Reading the file alone is not enough, because problems show up when text and speech happen together. For a 10 minute talk, a review pass usually takes far less time than translating from scratch.

You can do the whole job without paying:

If your video is already in Clipzi, you can skip the export, translate, re-import loop. On the Subtitles screen you first review the transcript against the video: fix words and whole lines, rename or merge speakers. Then, on paid plans, you translate the subtitles to another language and fix the translation line by line while you watch, with the original line next to it.

When it reads right, export one SRT for the whole video, or one SRT per speaker if you edit in Premiere (each speaker becomes its own caption track). The steps to bring those files in are in how to import SRT into Premiere Pro, and the speaker workflow is in subtitles by speaker. The Free plan includes 2 long videos, no credit card, so you can test the review flow before paying.