Export SRT and a burned-in MP4

Burn the short-form MP4. Export an SRT sidecar from the same transcript. Two files, two jobs. The muted feed wants pixels in the picture. The course player, the long-form YouTube master, and the client who will restyle language want a track they can switch off. Do not ship one object to both rooms and hope the player guesses.

This is an export procedure. It is not a lecture on which format is “better.” Burned-in captions and SRT are different artefacts. You already decided you need both. The work is: lock the cut, transcribe once, edit the words once, then render two deliverables from that file.

One transcript, two files

The source of truth is the timed transcript, not the MP4 and not the .srt. If you caption the social cut first and then try to “extract” an SRT from the burned file, you are reverse-engineering pixels. If you write an SRT first and then hope TikTok will draw it, you captioned a player that is not there.

Order:

Picture lock. Duration, ratio, and audio are the ones you will ship. A later trim is a new timing pass. Caption timing does not survive a recut for free.
Transcribe the locked file. On Versely that is add_veed_captions behind add captions to a video or the AI caption generator: 165 language codes. Pick the language that is actually spoken.
Edit the transcript as copy. Names, SKUs, prices, and the one claim the hook depends on. ASR misses those first. Spoken “um” does not have to become type. A marketing sentence the speaker never said is a text overlay, not a caption.
Freeze that transcript. That file is now the parent of both exports.

From here you branch. The burn-in path consumes the transcript plus a look, and composites type into frames. The sidecar path consumes the same words and times, and writes a text file a player can draw. They share a parent. They do not share an output.

Do not maintain two transcripts. A founder name fixed in the SRT and left wrong in the burn is two products. A cue split for reading speed on the social file and left as a paragraph in the sidecar is allowed — line breaks can differ — but the words and the clock should still match. If the voice says twelve days, both files say twelve days.

If you already have WebVTT from an editor or a platform download, do not rename the extension to .srt. The timestamp separator is enough for a player to reject the track: VTT uses a full stop in the fractional second, SRT uses a comma. Convert. The browser tool is VTT to SRT. Cue settings, STYLE, REGION, and inline classes do not survive, because SubRip has no representation for them. Timing and text do. That is the whole conversion.

If you already have SRT and the destination wants VTT, convert the other way. Do not paste a WEBVTT header onto a comma-decimal file and call it done.

Burn-in is the short-form MP4

Reels, TikTok, Shorts, paid social, and most UGC ads play muted until someone taps. The first sentence has to be in the picture. A .srt sitting next to the MP4 does nothing in that environment. The platform encodes the video. It does not look for your SubRip file.

Burn-in is that object: styled, timed words composited onto the frames. The viewer cannot hide them. The typeface is yours. Placement is yours, inside the safe area, not under the username stack and not under YouTube’s Shorts AI overlay. That is why the look is part of the edit rather than an accessibility toggle.

On Versely captions, look is also a bill:

21 BASIC presets bill at 1×. Legible, mostly still. Default for explainers and anything long enough that motion in the type would double the caption line for no reason.
9 DYNAMIC presets bill at 2×. Motion in the type. Default for a hook you want the muted feed to feel, not merely read.

There is no third tier. Preview a few seconds of this clip before you render the full file. Caption Studio’s 45 CSS presets, 9 groups × 5 variants, are a different catalogue used by the studio surface. Do not confuse those 45 with the 21+9 VEED looks. Pick one family for the channel and stay there. Caption styles is the look index.

Placement rules for the burned file:

Crop and lock ratio before you burn. Type in the lower third of a 16:9 frame will sit in the middle of a 9:16 frame, or vanish into the cut.
Keep the block off the extreme bottom and the button rail. Chrome moves. A percentage copied last year is a snapshot, not a contract.
Do not cover the mouth on a talking close-up. Do not cover the pack label the shot exists to show.
Watch the export muted, on a phone, once. If the first line sits under chrome, you did not finish.

The output is an MP4. That is the file you upload to the feed. Do not also attach the SRT to the TikTok “to be safe.” The feed will not draw it, and a reviewer will think you captioned the cut when you only captioned the archive.

Burn-in is pixels. You cannot switch it off later without a second job. Translation of a burned social cut is a new burn in the new language, not a second sidecar on the same picture. If this week’s slate is eight vertical ads, you owe eight burned files. You do not owe eight SRTs unless someone asked.

SRT is the long-form sidecar

A module, a long-form YouTube video, a broadcast master, a client who will translate, a player with a CC button: those want closed captions as a track. SRT is the interchange: numbered cues, start and end times with a comma decimal, plain text. No styling. No position. That is the feature. The player draws the words. The viewer can switch them off. A second language is a second file, not a second encode of the picture.

Burning English into a forty-minute course so that next quarter’s Spanish version requires a full re-composite is how teams learn this the expensive way. The picture should stay a picture. The words should stay text.

What belongs in the sidecar:

Speech, edited the same way you edited the burn: names right, filler optional, no invented marketing line.
Speaker IDs if the player needs them and the brief asked. Write them into the text (ALEX:) — SRT has no speaker field.
SDH — [door slams], [music] — only if the brief is SDH. Auto-transcribe will not invent those cues. Someone still authors them.

What does not belong in the sidecar:

A font. SRT cannot store one.
A raise to clear TikTok chrome. SRT cannot store position. If you need placement, stay in WebVTT until the player that will draw the captions, or burn the social file separately.
Karaoke / word-highlight. That is a burned-in look. It needs word-level timestamps in a compositor, not a SubRip cue.

YouTube long-form is the mixed case. Closed captions help search and accessibility. Burn-in on a talking-head YouTube cut is optional and often noisy. If the same story will also be clipped to Shorts, you still have two exports: a master with a sidecar, and vertical clips with burn-in. The transcript is shared. The files are not.

A course player that accepts VTT can keep cue settings the SRT conversion would strip. Ask what the LMS actually ingests. If it wants SRT, accept default placement. If it wants VTT, export VTT. Convert rather than guess.

Convert, then ship both

The last mile is packaging, not taste.

Social package. Burned MP4. Ratio locked. Safe area checked. Names spot-checked on mute. No sidecar required unless a paid-social spec asked for one as an archive. Upload the MP4.

Long-form package. Clean picture (no burn, or a burn you were explicitly asked to keep). SRT or VTT beside it. Same words, same clock. If the editor gave you VTT and the client asked for SRT, run VTT to SRT in the browser. Check the first and last thirty seconds after conversion. A whole track that is uniformly late is an offset. A track that drifts is a frame-rate mismatch; an offset will not fix that.

Client package that asked for both. Two files, labelled. spot_tiktok_burnin.mp4 and spot_master.srt (plus the unburned master). Put the transcript in the folder so the next language starts from words, not from a transcribe of a transcribe.

Do not:

Rename .vtt to .srt.
Burn first “to be safe” and then try to recover a clean master. You can transcribe again. You cannot un-composite type without covering it or cutting around it.
Send the burned MP4 to the LMS and the SRT to TikTok. That is the same two objects, swapped.
Treat platform auto-captions as either deliverable. YouTube may auto-caption. TikTok may draw a default. Those are the platform’s type, the platform’s timing, the platform’s spelling of your product name.

Caption cost on Versely is the burn, not the sidecar conversion. BASIC is 1×, DYNAMIC is 2×, for identical transcription work. What captions cost across a month of clips is that multiplier at cadence. The free converter does not caption the video and does not bill credits. Do not pay a generate to turn VTT into SRT.

If you need the studio to transcribe, burn the feed cut, and keep the transcript for the sidecar, sign up. Export the object each destination can actually play. A caption the player never draws is not a caption. Type welded into a course master is not a track.

FAQ

Can I upload an SRT to TikTok instead of burning?

Do not plan on it. The feed is a muted autoplay surface. Your SubRip file is not the soundtrack. Burn the line into the MP4 if the first sentence has to be readable with sound off. Keep the SRT for the player that has a CC button.

Do I transcribe twice if I need both files?

No. Transcribe once. Edit once. Branch. The burn consumes the transcript plus a look. The sidecar consumes the same words and times. Two transcripts is how a SKU is spelled two ways.

Why not just burn everything so I only have one file?

Because a burned course cannot hide the type, cannot restyle it, and cannot change language without a new composite. Short-form wants pixels. Long-form wants a track. One file asked to do both jobs fails in both rooms.