Add timed text you wrote, not ASR

Write each line and its clock. Add timed text overlays to a video is add_timestamped_captions: your copy, start_sec / end_sec per line. It does not listen. Auto-captions transcribe speech. Sending a mute pack demo to ASR will not typeset “12g protein” unless someone said it. Type the claim. Time it to the verb.

This is authored, sequenced type. It is not burned-in speech captions, and it is not one persistent badge for the whole clip.

You write every line and every clock

A sequence is the point: a question at 0s, a stat at 3s, a punchline at 7s. Each object needs text plus a start and end in seconds from the start of this file.

Procedure:

1. Picture lock. Timestamps on a file you will trim are a recut of every cue.
2. Crop to the ship ratio first. Type placed on 16:9 will sit wrong on 9:16. Aspect, then overlays.
3. Script beats as short lines. One claim per beat. A paragraph is a caption track you did not transcribe.
4. Watch once with a timer. Mark in and out against the picture, not against a song you have not attached yet. If the bed is coming, attach or mix it first so you are not timing to silence you will cover — or accept that type is timed to picture, not to the snare.
5. List the cues. No gaps that look like a missed beat if the joke needs a hold; gaps are allowed when the picture must breathe.
6. Set size, colour, font, position once for the whole set. There is no per-line colour. If beat three must be red and the others white, that is two renders or a different tool.
7. Render. Watch muted. Every line must be true of the frames it sits on.

Precision is yours. The tool will not detect beats. It will not karaoke a word you did not type. list_caption_fonts if the typeface is a named brand font rather than a default.

Cost is a standard editor render, same family as a single overlay. estimate_cost if you want the number.

Illegal uses:

• Dumping an ASR transcript into timed overlays because you wanted to restyle cues by hand. If the source is speech, use the caption tool and a preset. You will mistime a 40-cue dump.
• Timing a legal disclaimer to flash at 0.3s. If it must be read, it must be holdable. Subtitle reading speed exists for a reason; authored claims have the same eyes.

This is not a caption track

Add captions to a video (add_veed_captions) transcribes the soundtrack, times to the mouth, burns a BASIC or DYNAMIC look. 165 language codes. 21 BASIC at 1×, 9 DYNAMIC at 2×. That is speech.

Timed overlays are copy you authored. Nobody had to say it. Mixing the two jobs is how a CTA you typed gets treated as a subtitle, or a spoken line gets treated as a poster.

When to use which:

• Talking-head, the words matter on mute → captions (ASR).
• Silent pack, the claim is a number or a hook you wrote → timed overlays (or one persistent overlay if it never changes).
• Both → layout. ASR in the caption band, authored beats higher or lower on purpose, not two karaoke stacks. Safe area is already crowded on 9:16.

Burned-in captions are still the short-form speech object. Timed overlays are not “custom captions.” They do not get word-highlight from the engine. If you want karaoke of speech, that is DYNAMIC captions, not this tool.

Sidecar SRT is a third object: a file, not pixels. Burn-in is what the feed sees. Do not upload an SRT of authored marketing claims and expect TikTok to typeset them as a hook. Platforms treat sidecars as subtitles of speech.

Caption timing still applies as a reading problem. A 1.5-second flash of a 12-word claim is not a claim. Hold it until a mute scroller can finish it.

One look for the set

Font, size, colour, position are global for the render. That is a limit. Use it as a brand constraint. A rainbow of per-beat styles is a different editor.

Position is a band (top / center / bottom), not a free pixel. Centre on a pack shot covers the SKU. Bottom on 9:16 hits chrome. Top is often the hook band. Pick once, preview on the real background.

Contrast: white on a white pour fails. A background block helps. Preview. preview_caption_style is for VEED speech presets; for this tool, render and look. Do not guess on a black slate if the video is a sunlit table.

If you also burn ASR, do not put timed overlays in the same band. Two clocks, one strip: both lose. Author the stack on paper.

YouTube Shorts photoreal overlay (YouTube Blog, 27 May 2026) is more furniture on the picture. Long-form labels sit below the player. Timed hooks on a Short need room at the top.

Article 50 (2 August 2026) does not care that your overlay is “just text.” The picture can still be photoreal AI. Type is not a wash.

Place after picture lock and after the crop

Order that avoids paying twice:

1. Picture lock.
2. Ratio (9:16 / 1:1 / 16:9).
3. Audio lock (VO, dub, bed) if the beats should hit words. If beats hit picture, audio can wait.
4. Timed overlays.
5. ASR captions if speech also needs a transcript, in a different band.
6. Export.

If you localise, authored claims must be rewritten, not transcribed. French ASR of an English VO will not produce your French hook. Translate the overlay list. New clocks if the line lengths change.

A single persistent badge — “Recipe” for the whole clip — is add text overlay, simpler. Timed overlays are for changing lines. Do not use this tool for one string that never leaves.

Do not generate the beats into the still or into a T2V prompt. Model-painted type is misspelled texture you cannot retime.

Write the beats in Versely with timestamps on the locked file. Do not send a mute clip to ASR and hope it invents the claim.

FAQ

Does this transcribe the video?

No. You write and time every line. For auto-transcribed subtitles, use add-captions-to-video.

How is this different from one text overlay?

One overlay is a single line for the whole clip. Timed overlays are multiple lines, each with its own in and out. Same authored-copy family. Different clock.

Can each line be a different colour?

Not in one render. Size, colour, position, and font are set once for the set. Split renders if a beat must break the look.