A video caption generator burns in
A video caption generator composites words into pixels. That is burn-in, not a sidecar SRT the feed will never draw.
A video caption generator burns in. It transcribes speech, times the lines, and composites styled text into the frames. The words become pixels. They cannot be switched off, restyled by Instagram, or lost when you re-upload. That is the object behind the query "video caption generator." It is not a sidecar .srt.
Search Console still shows that string (13 impressions, seven days to 4 September 2026) next to "ai caption generator" and "auto captions." The matching tool is the AI caption generator. This page is the carriage: burn-in versus sidecar, so you stop shipping a silent MP4 plus a text file to TikTok.
Burn-in is a render, sidecar is a track
Burned-in captions are open captions: the text is in the picture. Download the file, hand it to a platform with no caption engine, and the words are still there. The look is yours — font, colour, animation, placement — not the platform's default CC skin. For a muted autoplay feed, both properties are mandatory. The viewer will not enable a track. The caption is part of the edit.
A sidecar is closed captions: SubRip (.srt), WebVTT (.vtt), or a platform-native asset. The player draws it. The viewer can turn it off. YouTube long-form, a course <track>, a broadcast caption encoder — those want a sidecar, often with SDH content (speech plus non-speech plus speaker IDs). The feed does not.
The generator on Versely is a burn-in tool. Styled, auto-transcribed captions are composited into the MP4. It is the right object for Reels, TikTok, Shorts, and ads that re-encode and strip tracks. It is the wrong object if the brief was "upload a VTT to the LMS." For that, take a plain transcript, author the file, and keep the picture clean — or ship both: a burned short-form cut and a sidecar for the archive.
Renaming .srt to .vtt is not a generator. The timestamp separator and the missing WEBVTT header are enough for a player to reject the track. In-browser converters exist on free-tools for that mechanical job. They do not caption the video. A script timer does not caption the video either; it divides word count by WPM.
Karaoke / word-highlight families need word-level timestamps and a line of actual speech. They are still burn-in. They are not a sidecar feature. SRT cannot represent a highlight moving from word to word. If the client asked for "animated SRT," they asked for a burned MP4 plus a plain .srt for archives, or they meant WebVTT and have the wrong extension in the brief.
What the generator actually does, in order
Lock the cut. Pixels do not ripple. A later trim is a new caption render, not a nudge of a cue.
Transcribe. Language code = language spoken. The tool does not translate. 165 language codes live on the add captions map. Wrong code = phonetic garbage.
Read the transcript. Names, prices, the one claim the hook depends on. Auto captions miss those first. A wrong word on a mute feed is the whole message.
Split lines. Latin-script captions are commonly kept near 42 characters and two lines. A paragraph on a face is not a caption.
Pick a style on this footage. Caption styles is the catalogue. 21 basic looks at 1× the caption render, 9 animated looks at 2×, same transcription. Preview a few seconds; the 480p preview pass is free with a short per-user cooldown. Do not caption three full videos to decide a look. What captions cost across a month of clips is that multiplier at cadence.
Place off the chrome. Safe area is the part of 9:16 the UI does not cover. The bottom strip is account name, caption field, buttons. A generator that defaults to "bottom centre" will lose the line in the feed. Check a phone preview.
Watch on mute. Caption timing errors show as late lines, hung lines, or a last word that vanishes. Reading speed errors show as a line you cannot finish. Fix by splitting or holding, not by picking a punchier preset.
Export the MP4. That file is the captioned video. Do not also upload an SRT to Instagram and expect a second, better caption. The feed will play the picture.
Overlay is a sibling, not a substitute. A hook you wrote can sit above the transcript captions. If both say the same sentence, delete one.
When you still need a sidecar
YouTube long-form: a closed track the host can toggle, translate, and index. Burn-in as the only captions kills off-switch and makes a second language a full re-render.
Courses and HTML5 players: kind="captions" WebVTT, labelled. SDH if the brief is accessibility rather than "subs for noisy gyms."
US linear: 608/708 inside the essence. A generator burn is not that package. A caption house maps a timed-text master into 608/708. If your master was speech-only, the packets will be speech-only and can still fail completeness under 47 CFR 79.1.
Localisation: one picture, several VTTs. Burn-in per language only when the destination is a feed that will not draw a track.
Forced narrative: a small file of signs and foreign lines, often burned because the plot depends on it staying on with CC off. A full-dialogue generator export is the opposite of forced. Author the small file. Forced narrative is the definition.
If a client wants both, render the burned short and export the transcript as SRT or VTT. Say which file is for which destination. CC vs subtitles vs forced narrative is the memo language.
Article 50 (2 August 2026) is a label on certain synthetic media, not a caption format. Do not hide the EU mark inside a cue and call the generator a compliance tool.
The test
Will this file play in a muted feed with no CC button? Burn in. That is a video caption generator.
Will a player draw a track the viewer can toggle? Sidecar. The generator can still give you a transcript; it should not be the only delivery.
Did you search "video caption generator" hoping for an SRT download as the finished job for TikTok? You will upload a silent picture. Change the object.
Did you search it hoping for a new video with words invented in the pixels by a T2V model? That is typesetting inside a generate, and it is a bad caption. Generate the clip, then burn.
A video caption generator is a burn-in. Keep the sidecar for the rooms that have a CC button.
Burn the locked cut in the caption generator. Sign up when the speech already exists and the next file should be an MP4 with the words in the picture.
FAQ
Does a video caption generator give me an SRT?
Not as the finished short-form object. Versely's generator burns into the MP4. You can take a transcript for a sidecar when a player needs one. TikTok and Reels will not play that sidecar as the caption.
Why not always use a sidecar so I can restyle later?
Because muted feeds will not enable it, and ad platforms often strip tracks. Restyling a burn is a new render. That cost is the price of a caption that actually appears.
Are burned-in captions the same as closed captions?
No. Closed captions are a switchable track. Burn-in is open: pixels, always on. Short-form wants open. Long-form players want closed, often with SDH content in the file.