Workflow guide
Captions people actually read instead of skimming past
Most viewers watch with sound off at least some of the time, and platforms rank captioned video as more accessible. Good captions are timed to speech, broken into readable lines, and styled so they never fight the picture. This guide covers the full pass from transcript to export.
From the Mellius product team · Updated September 28, 2026
Conspiracy theorists, you may hate this
AI Director
›
Just type what you need
- ›Remove all silent parts
- ›Add bold kinetic captions
- ›Only have Bill Ackman speaking
- ›Cut to highlights only
- ›Add cinematic transitions
- ›Match cuts to the beat
The short answer
Transcribe the audio, correct names and terms, split captions into one idea per card of 32 to 42 characters, keep each card on screen one to six seconds, style with high-contrast text and a soft background, then export burned-in for social and sidecar SRT for YouTube.
Timing matches speech
Each caption appears within a few frames of the word being spoken and clears before the next idea starts.
Readable at a glance
Short lines, large type, and strong contrast let viewers read while still watching the picture.
Export fits the platform
Burned-in captions for feeds that mute or crop, plus a sidecar file where the platform supports uploads.
Get an accurate transcript first
Auto-transcription does the heavy lifting, but proper nouns, brand terms, and homophones still need a human pass. Read the transcript against the audio at 1.25x speed and fix names, numbers, and domain words first, because every caption downstream inherits these errors. In a traditional editor this means the Captions panel in Premiere or the Subtitles inspector in DaVinci Resolve; in Mellius you can ask for a transcript and then correct lines conversationally before generating captions.
Decide between verbatim and clean-read captions. Verbatim keeps every filler word, which suits legal or accessibility deliverables. Clean-read removes ums, false starts, and repeated words, which suits marketing and social video where reading speed matters. Pick one style per project and stick to it so the captions feel consistent.
- Fix names, places, and product terms before touching timing.
- Keep one idea per caption card; split long sentences at clauses.
- Aim for 32–42 characters per line and at most two lines.
- Remove filler words for social; keep them for accessibility masters.
Time, break, and style the lines
Set caption duration between one and six seconds per card and never let two speakers share one card without a dash or line break to separate them. Snap caption boundaries to pauses in the waveform rather than to fixed intervals, leaving at least two frames of breathing room across shot cuts so the text does not visibly pop. Reading speed around 160–180 words per minute is comfortable; faster than that and viewers stop watching the picture.
Style for contrast in every scene: white or cream type with a soft dark box or stroke, positioned in the lower third but raised above platform UI overlays on 9:16 video. Test on a phone at arm's length — if you cannot read it over the brightest shot, increase the background opacity rather than the font size. Common mistake: animating every word with bouncy presets, which looks energetic for three seconds and exhausting for three minutes.
- Keep captions clear of faces, lower-thirds, and platform buttons.
- Use sentence case, not all caps, for body captions.
- Reserve kinetic word-by-word styling for hooks, not full videos.
- Proof captions on mute — they should carry the story alone.
Export for each destination
YouTube, Facebook, and LinkedIn accept sidecar SRT or VTT files, which keeps text crisp and lets viewers toggle captions. TikTok, Reels, and Shorts need burned-in captions because viewers cannot toggle sidecars and cropping can bury lower-third text. Export a master with a sidecar plus separate burned-in vertical renders rather than forcing one file to serve everywhere.
Before publishing, watch the export on the smallest screen it will meet and check three things: sync against speech, spelling of every proper noun, and overlap with on-screen graphics. In Mellius, captions live as editable timeline objects, so you can ask to restyle, reposition, or re-export them without rebuilding the project, and undo returns the previous version.
Compare the workflows
| Method | Best fit | Trade-off |
|---|---|---|
| Manual captioning | Maximum control over timing and line breaks | Slow on anything longer than a few minutes |
| Auto-transcribe then correct | Most talking-head and interview work | Needs a careful proper-noun pass |
| Template or preset styling | High-volume social output | Can look generic without custom tuning |
A practical sequence
Put it into practice
- 01
Transcribe and clean
Generate a transcript, fix names and terms, and choose verbatim or clean-read style.
- 02
Time and style
Break lines at 32–42 characters, sync to speech, and apply one high-contrast style.
- 03
Export per platform
Ship sidecar SRT/VTT where supported and burned-in renders for vertical feeds.
Frequently asked questions
Should captions be burned in or uploaded as SRT?
Use sidecar SRT or VTT on YouTube, Facebook, and LinkedIn so text stays sharp and toggleable. Burn captions into TikTok, Reels, and Shorts, where sidecars are unsupported and autoplay is muted. Many creators keep both: a sidecar master plus burned-in vertical cuts.
How long should each caption stay on screen?
Keep most cards between one and six seconds, paced to speech rather than fixed intervals. Very short cards flash by unread, and very long ones desync from the voice. If a sentence needs more than six seconds, split it at a clause boundary.
What font size works for captions?
Size is relative to frame, but a good rule is legible on a phone at arm's length over your brightest shot. Prioritize contrast — cream type on a soft dark box — over sheer size, and keep captions inside title-safe margins so platform UI never covers them.
Continue reading
Primary sources
Product capabilities change. These official references support the comparisons above.
Describe the edit. Keep the timeline.
Mellius turns conversational direction into cuts, captions, graphics, motion, color, and sound you can still inspect and change.
Start editing free