Caption the whole cut in the agent

Caption the whole cut by asking the Versely agent to run transcribe and caption my video. That is the named job: it transcribes the speech on the file and burns styled, word-timed captions with VEED subtitle presets. It is not a sidecar exporter, not a headline tool, and not a translator. Give it a locked cut with audible speech, a preset, and a language code.

The same render also lives on add captions to a video. The agent page is the chat door; the editing page is the task door. They call add_veed_captions. Pick one door per session. Do not caption the same file twice because you opened both.

The job is transcribe-and-caption-my-video

The slug is the contract. transcribe-and-caption-my-video takes a video that contains speech and returns that video with captions composited into the frames. Supporting tools on the same capability are preview_caption_style and list_caption_fonts. That is the whole surface.

What you say to the agent can be ordinary:

• Add captions to this video in the "glass" style
• Caption this in Spanish with the "whisper" preset
• Preview the "fusion" caption style on the first few seconds before you caption the whole thing

If you do not name a vibe, the job defaults to Glass. Glass is a Dynamic preset. Dynamic presets bill at 2× the Basic rate for the same transcription. The default is how a 2× month happens without anyone choosing a look. Name a Basic preset when you want 1×. Name Glass when you actually want Glass.

The video must contain audible speech. A silent product plate, a screen recording with no VO, or a clip that is only a music bed will give the transcriber nothing useful. The agent will not invent a script you did not say. If the hook is a line you wrote and nobody spoke — "WAIT FOR IT", a price, a URL — that is a different capability.

Language is a code, not a hope. The job transcribes in 165 language codes. It transcribes what is spoken. A French VO with en-US still on the request is a bad English transcript. A dubbed cut is a new caption job after the dub, in the new language. There is a separate translate task for changing the language of the video; do not ask the caption job to do that in one pass.

Output is an MP4 with the captions burned in. There is no cue list to nudge later. A timing fix is a new render. That is why the cut has to be locked, and why the preview exists.

What this job will not do

Three neighbour capabilities sit next to it in the editor agent. Mixing them is the usual failure.

Add a text overlay to my video burns one authored line — a hook, a headline, a CTA — for a stretch of the clip. It does not listen to the audio. Use it when the words are yours. Do not use the caption job hoping a headline appears; it will typeset the VO instead, or typeset nothing.

Add timed text overlays to my video burns several authored lines, each with its own start and end. You supply the copy and the clock. Nothing is transcribed. Use it for "Day 1 / Day 30 / results" chapters, not for dialogue.

Build a video from clips, music, and captions is an assembly job. It wants clips, a transition, a music track, and caption text you already have. It is not the transcribe-and-burn path for a finished talking-head.

The caption job also will not:

• Place closed captions as a .srt or .vtt sidecar. It burns pixels. If you need a track for YouTube long-form or a course player, transcribe to text separately and export a sidecar. That is a different object.
• Restyle a burned file without re-hearing the audio. There is no "just change the font" pass on pixels. You recaption the source.
• Fix a cut that is still moving. Trim first. Caption second. A bumper you drop after burn-in takes every cue with it.

If you need fonts enumerated before you pick a look, list_caption_fonts is the supporting tool. If you need to see the look on this clip, preview. Then caption the whole cut once.

Preview, language, 1× / 2×

Three decisions belong in the prompt before the full render.

Preview. Ask the agent to preview a named preset on the first few seconds of this file. Contrast against this background, this face, this motion. A studio sample of Glass on a different talker is not a preview. Two candidate presets are enough. Captioning the whole cut in three looks to pick one is the expensive way to decide.

Language. Set the code to the language actually spoken. Accents inside a language are a transcription-quality problem, not a different job. A second language on the same tape — a guest, a quote, a product name in English inside a Spanish VO — is where auto captions miss first. Read those strings after the burn, or transcribe to text and correct them before you commit a style you like.

Tier. Basic (21 looks: Simple, Plain, Beans, Corpo, and the rest) bills at 1×. Dynamic (9 looks: Glass, Whisper, Glide, Fusion, Terminal, Handwritten, Backdrop, plus the remaining premium set) bills at 2×. What captions cost across a month of clips is that multiplier at cadence. Transcription is the same work. The animation is what you pay extra for. Pin 1× for readable talking-heads. Spend 2× when the caption is the performance.

Caption styles is a separate catalogue: nine families of five live presets (Classic, Paper, Ink, Halo, Hype, Signature, Headline, Snap, Reveal). VEED Basic/Dynamic is what this agent job actually burns. They are not the same list. Classic is not Corpo. Snap is not Glass. Ask for a VEED preset name the job knows, or you will get the Glass default.

Cost is quoted before the render. Do not invent a credit figure in the prompt or in a client email. Read the quote. estimate_cost exists if you want the number on a long file before you commit.

Caption the locked cut, then stop

Order of operations is the rest of the craft.

1. Lock picture and audio. No pending bumper, no VO pickup in Slack.
2. Confirm there is speech. If there is not, stop and use an overlay job or add a VO.
3. Preview one or two presets on the first seconds.
4. Name preset, language, and placement in the prompt.
5. Run transcribe-and-caption-my-video on that file.
6. Watch the export on mute, on a phone. Read the names, the numbers, the first cue. Check the safe area.
7. If a word is wrong or a cue is late, recaption the source. Do not "nudge" pixels.

A second agent turn that says "make the captions bigger" on a burned MP4 is a new full job, not a tweak. Say what you want in the first prompt. If the transcript is the risk (legal names, a price, a claim), pull text first with the transcript task, correct it, then caption. The agent will not know that your brand is not the homophone it heard.

The job is built for the whole cut you intend to publish — a 15s ad, a 45s testimonial, a Short. It is not built for a two-hour interview you have not clipped. Trim to the deliverable, then ask.

Once the file is right, you are done. Do not also attach a sidecar and assume TikTok will draw it. Do not run the overlay job on top unless you have a second, authored line that the transcript should not carry. One speech track, one caption burn, optional one hook overlay. That is a finished cut.

Caption the locked cut in Versely.

FAQ

Is transcribe-and-caption-my-video the same as add-a-text-overlay-to-my-video?

No. The caption job transcribes speech and times it. The overlay job burns a line you wrote and never listens to the tape. Use both only when you need spoken captions and a separate headline.

What happens if I forget to name a preset?

The job defaults to Glass, which is a 2× Dynamic look. Name a Basic preset if you wanted 1×. Preview first if you are not sure the look fits this clip.

Can the agent caption a video with no one talking?

Not usefully. It transcribes real speech. A silent clip needs an authored overlay, a VO, or a different edit. Do not run the caption job to see what happens.