To create your own subtitles without an AI tool, you need four things: the words, a rule for splitting them into short cues, a plain text editor, and a player to test in. The file you make is an SRT. It is a numbered list of cues, each with a start time, an end time, one or two lines of text, and a blank line. No special software is needed to write one, and VLC or YouTube Studio will tell you within a minute whether it works.
Here is how to create your own subtitles by hand, in six steps: write the lines, split them into cues under the line-length and reading-speed limits Netflix publishes for its subtitlers, type the SRT with a three-cue worked example, time it, test it in VLC and YouTube Studio, and burn it in if the video is going to a social feed. The last section is honest about when the manual route is worth the hours and when an automatic draft plus an edit gets you the same file faster.
Step 1: write the lines
Start with the words, not the times. Play the video and type what is said, one sentence per line, in a plain text file. A two-minute clip has about 250 to 350 spoken words, which takes most people 15 to 25 minutes at this stage.
Leave out fillers and false starts ("um", "like", a sentence the speaker abandons and restarts) and repeats, so "I think, I think it works" becomes "I think it works". Subtitles are read, not heard, and a reader does not need the stumble.
Keep three things:
- Speaker changes. When two people speak inside one cue, Netflix's English Timed Text Style Guide says to use a hyphen with no space before each speaker's line, one speaker per line.
- Sounds that matter to the story. The same guide uses square brackets for sound effects in subtitles for the deaf and hard of hearing, in lowercase except for proper nouns, and only when the sound cannot be seen. A phone ringing off screen gets [phone rings]. A visible laugh does not need [laughs].
- Exact wording. Brand names, code, a legal line, a joke that depends on one word: you type it exactly, once.
Punctuation and capitalization rules are in subtitle punctuation and capitalization, so this post does not repeat them.
Step 2: split the lines into cues
A cue is one block of text that sits on screen for a set time. Two limits decide how to split your lines, and both come from a published style guide rather than taste. The figures below are from Netflix's English style guide, read 28 October 2026.
| Limit | Netflix English Timed Text Style Guide |
|---|---|
| Characters per line | 42 |
| Lines per cue | Maximum two |
| Reading speed, adult programs | Up to 20 characters per second |
| Reading speed, children's programs | Up to 17 characters per second |
Reading speed is the one people skip. Characters per second is the total characters in the cue, spaces included, divided by how long the cue is on screen. A 60-character cue needs at least three seconds at 20 characters per second. If the speaker says it in two, you cut words or let the cue end a little after the speech does.
Where to break a line, from the same guide:
- Keep the text on one line unless it is longer than 42 characters.
- When you must break, make the bottom line the longer one, but do not leave one or two words alone on top.
- Break after punctuation, before a conjunction such as "and" or "but", and before a preposition such as "on" or "with".
- Do not split a noun from its article or adjective, a first name from a last name, or a verb from its subject pronoun or its auxiliary.
So "Today we are fixing the noisy fan on my old laptop." is 50 characters and needs two lines. Break before "on". Breaking after "noisy" would separate the adjective from its noun.
Step 3: write the SRT by hand
Open a plain text editor: Notepad on Windows, TextEdit on a Mac set to plain text, or any code editor. Not Word, which adds curly quotes and its own file format. What is an SRT file covers every byte of the format, so here is only what you type.
Each cue is four parts: a counter, a timing line, the text, and a blank line. The timing line is hours, minutes, seconds, a comma, three digits of milliseconds, a space, an arrow made of two hyphens and a greater-than sign, a space, and the end time in the same shape. Here is a complete three-cue file for the opening ten seconds of a laptop repair clip. The right-hand column is exactly what goes on each line.
| File line | What you type |
|---|---|
| 1 | 1 |
| 2 | 00:00:01,200 --> 00:00:03,900 |
| 3 | Welcome back to the channel. |
| 4 | (empty line) |
| 5 | 2 |
| 6 | 00:00:04,100 --> 00:00:07,600 |
| 7 | Today we are fixing the noisy fan |
| 8 | on my old laptop. |
| 9 | (empty line) |
| 10 | 3 |
| 11 | 00:00:07,800 --> 00:00:10,300 |
| 12 | -You ready? |
| 13 | -[sighs] Let's go. |
| 14 | (empty line) |
Four things to check before you save:
- Commas, not periods, before the milliseconds. 00:00:01,200 is SRT. 00:00:01.200 is WebVTT, and many players will reject the file or ignore the cue.
- Every time has the full shape. Two digits for hours, minutes and seconds, three for milliseconds. 0:00:01,2 fails strict parsers.
- One empty line after every cue, including the last. The blank line is how the player knows the cue has ended.
- Save as .srt in UTF-8. In Notepad, choose All Files as the type, name it clip.srt, and pick UTF-8 in the encoding menu. YouTube's supported subtitle files page says the file "must be in plain UTF-8", and the same encoding keeps accents and non-Latin scripts intact in every player.
The example checked against the limits
Count the characters in each cue, spaces included, and divide by the cue's time on screen.
| Cue | Characters | On screen | Characters per second | Longest line | Passes 42 and 20 |
|---|---|---|---|---|---|
| 1 | 28 | 2.7 s | 10.4 | 28 | Yes |
| 2 | 50 | 3.5 s | 14.3 | 33 | Yes |
| 3 | 29 | 2.5 s | 11.6 | 18 | Yes |
Now the failing version. Suppose you had kept cues 2 and 3 as one block: "Today we are fixing the noisy fan on my old laptop. You ready? Let's go." That is 72 characters. Stretched across the full 6.2 seconds from 4.1 to 10.3 it is 11.6 characters per second, which passes, but it cannot fit two lines of 42 without a third line, which fails. And if the speaker had said all of it in 3.5 seconds, the single cue would be 20.6 characters per second, over the limit. Splitting at the speaker change fixes both at once. That is the whole craft: split where the speech splits, then check the numbers.
The subtitle reading speed checker does this arithmetic for a whole file, in the browser without uploading it, and flags every cue over the characters-per-second limit you set.
Step 4: time the cues
Timing by hand means finding the start and end of each spoken line and writing the numbers down. Two ways:
- Scrub in a player. Play the clip, pause at the first word of each line, and read the time off the player. VLC shows elapsed time to the second; for the milliseconds, step frame by frame and estimate, or accept whole seconds for a first pass. Exact but slow: expect two to three minutes of work per minute of video.
- Start with even spacing and adjust. Paste your lines into the TXT to SRT generator, choose a seconds-per-cue value, and it writes sequential timestamps for every line and downloads an .srt, in the browser. The timings are evenly spaced, not matched to the speech, so the next pass is to move each start and end to where the words fall. Most of the numbers are already close, so this beats scrubbing from zero.
Starts are what viewers notice; an end that runs 200 milliseconds past the speech is invisible. Two working rules used in the example above, the author's rather than Netflix's: leave about 200 milliseconds between cues so the eye registers the change, and never hold a cue for less than one second.
Step 5: test in VLC and in YouTube Studio
Test before you publish, because an SRT that fails does so silently: the player shows nothing.
VLC. Name the subtitle file exactly as the video, with only the extension changed: laptop-fan.mp4 and laptop-fan.srt in the same folder. The VideoLAN wiki says that when both names match, VLC opens the subtitle file automatically with the video. Then:
- Play from the start. If nothing appears by the first cue's start time, press V, which the VLC hotkeys table lists as "Cycle subtitle track", and Shift+V to toggle subtitles on. If still nothing, the file is malformed: check the comma, the blank lines and the .srt extension.
- Watch for early or late cues. If every cue is off by the same amount, press H (subtitle delay up) or G (subtitle delay down) until they land, and note the delay VLC shows. That is your offset. Fix it in the file, because VLC's delay is not saved into the SRT.
- Watch for one cue that is wrong while the rest are fine. That is a typo in one timing line, usually a minute digit.
- Read every cue at normal speed without pausing. If you cannot finish one before it leaves, it is over the reading-speed limit, whatever the arithmetic said.
Loading an SRT through VLC's menus, and what to do when it will not load, is covered in can VLC auto-generate subtitles.
YouTube Studio. YouTube Help lists the steps: sign in to YouTube Studio, choose Subtitles in the left menu, select the video, click ADD LANGUAGE and pick the language, click ADD under Subtitles, choose Upload file, choose With timing, click Continue, pick your .srt, and Save. Studio then shows the cues against the video. Play it and check the same four things. If Studio rejects the file, the usual cause is a period in the milliseconds or a file saved as .txt. YouTube's formats page lists SubRip .srt among the basic formats it accepts, so a correct file goes through.
A hand-written SRT that passes both tests will play in almost any player. Keep the file; it is the master copy.
Step 6: burn it in for social feeds
An SRT is a separate track. YouTube and most players read it and let viewers turn it on or off. Instagram Reels, TikTok and most short-form feeds do not take a sidecar file for the main caption layer, and feeds autoplay on mute, so for those the text has to be drawn into the video frames.
Desktop tools burn an SRT into an MP4 for free: burn subtitles with HandBrake, VLC and FFmpeg has the exact settings, and the burn subtitles to video page covers the browser route.
CapsAI is a different workflow rather than a step in this one. It captions an uploaded MP4 or MOV up to 10 minutes long in one of 11 spoken languages, including English, Hindi and Hinglish, shows the timed draft for you to correct word by word, and exports one MP4 with the captions burned in. It does not import your SRT and does not export one. New accounts get 3 free minutes to start. Packs cost $4, $9 and $19 (₹49, ₹99 and ₹299 in India; €4, €9 and €19 in the euro area) for 20, 60 and 180 minutes, paid once, and watermark-free export comes with every paid pack.
When hand-made subtitles are worth it
Be honest about the time. A two-minute clip, done by hand with the steps above, is 45 to 90 minutes the first time and perhaps 30 once you have the rhythm. The result is exactly what you typed, with no misheard words to hunt for. It is worth it when:
- The clip is short. Under about three minutes, the hand route costs less than learning and correcting a tool.
- The language is not covered. Generators cover a fixed list of languages. If yours is not on it, or the clip mixes two languages a generator handles badly, typing is the only way to get it right.
- The wording must be exact. Scripts, product names, legal text, a line break that is part of the meaning.
An automatic draft plus an edit is faster when:
- The clip is long. At ten minutes, transcription alone is an hour of typing. A generator gives you timed text in minutes, and your job becomes reading it against the audio and fixing the misses.
- Timing is the hard part, not the words. Hand timing is the slowest step in this guide, and a draft already close to the speech removes most of it.
- You are making many clips. The limits in step 2 still apply, but you check them on a draft instead of a blank page.
For the automatic route, the auto subtitle generator covers how a draft is made and what to check in it. For the hand route, the TXT to SRT generator in step 4 gets you from a text file of lines to a timed draft you finish yourself. Either way, the file should pass the same two tests: it loads in VLC with the same name as the video, and every cue can be read before it leaves.
