How to get the transcript and subtitle files for your clips

Last updated 2026-09-01 · 7 min read

What is the short version?

Every clip you have already made contains a finished piece of writing that nobody has bothered to collect. It is the same words, in the same order, delivered well enough that you kept the take — and it is sitting inside an MP4 where no search engine and no reader can reach it.

1. Go to Vizard Agent and ask for the transcripts.
2. Say which clips, or say all of them.
3. Say whether you want captions, a readable draft, or both.

What do you need before you start?

The clips, and that is genuinely all. Vizard Agent works from the files already in your project, so there is nothing to upload again and no need to remember which version of which cut was the one you actually published.

• The clips. Delivered, or freshly uploaded.
• The purpose. Subtitles, reading, or search.
• The language. If it is not obvious from the audio.
• The order. If they belong together as chapters.
• The format. Document, subtitle file, or both.

What do you type into Vizard Agent?

You can be very casual about this one. It is one of the few requests where a half-sentence works, because there is no creative decision hiding inside it — Vizard Agent knows which files it delivered and what transcribing them means.

Prompt

Variants worth knowing:

• "Readable, not verbatim." Punctuated prose, stumbles removed.
• "Keep every word." Verbatim, for legal or archival use.
• "Join them into one document." Chapters instead of files.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real project that started as a question about transcripts and ended as a seven-chapter manuscript. Step nine is the one that separates a transcript from something you can actually read.

1. Checks your project files and history.
2. Checks the transcription tool's options.
3. Checks the length of the delivered videos.
4. Pulls the spoken transcript from the delivered cut.
5. Creates and uploads the transcript files.
6. Transcribes the newly uploaded clips and views frames from each.
7. Builds a readable document and a subtitle file for every clip.
8. Polishes the punctuation in the readable versions.
9. Scans the chapters for leftover speech artefacts.
10. Cleans them into a reading draft.
11. Checks how each cleaned chapter now opens.
12. Tracks down why doubled words survived the cleanup.
13. Clears the last stumbles and re-runs the check.
14. Assembles all seven pieces into one working manuscript.
15. Reads a full chapter back to confirm nothing was mangled.
16. Checks the chapter lengths are intact after cleaning.
17. Uploads the finished files.

Step twelve is the interesting failure. A first cleanup pass removed most of the repeated words and left some behind, and rather than running the same pass again and hoping, Vizard Agent worked out why the survivors survived — then fixed the rule rather than the symptom. Step sixteen is the safety check on that: cleaning prose is exactly the operation that quietly deletes a paragraph.

What does the result look like?

Two files per clip: a subtitle file that will load into any editor or platform, and a readable document with proper sentences, capital letters and full stops where the speech had pauses. Where the clips belonged together, one further document holds them all as ordered chapters.

The readable version is not the same text as the subtitle file, and it should not be. Captions follow speech; prose follows sentences. Vizard Agent produces both rather than making you choose which compromise to live with, and the two stay in step because they came out of the same word timings.

When does this not work well?

Transcription is only as good as the audio underneath it, and a readable draft asks for one more thing on top: that the speech was structured enough to become sentences. These are the cases where the output needs a human pass before it goes anywhere.

• Heavy background noise. Words are guessed, not heard.
• Several people talking over each other. Attribution gets muddled.
• Specialist terminology. Names and jargon need checking.
• Very heavy accents in a second language. Accuracy drops.
• Legal or medical use. Verbatim, and check it yourself.

How do you fix a result that came back wrong?

Name the term or the chapter. Vizard Agent keeps the word-level timings, the raw transcript and the cleaned draft as separate things, so a corrected spelling propagates into both the subtitle file and the prose without either being rebuilt from the audio again.

• "That name is spelt wrong." Corrected everywhere it appears.
• "Chapter three lost a paragraph." Restored from the raw transcript.
• "Keep the ums." Re-issued verbatim.
• "Wrong subtitle format." Re-exported; text unchanged.

How does Vizard Agent compare to doing it yourself?

By hand, transcription is either an afternoon with a foot pedal or a paid tool that gives you an unpunctuated wall of text you then have to edit into something a human would read. The second half — turning a transcript into prose — is the part no tool usually does.

Common questions

Does it cost a separate job? It is part of the same project. Ask and Vizard Agent produces them.

Can I get an SRT? Yes. Vizard Agent produces standard subtitle files you can load anywhere.

Is the readable version verbatim? No, by default. Say "keep every word" and Vizard Agent leaves the stumbles in.

Can it do several clips at once? Yes. Vizard Agent transcribes them in parallel and delivers a set.

Will it name the speakers? Ask and it will. Vizard Agent can separate voices where the audio allows.

What about a video I did not make here? Upload it. Vizard Agent transcribes uploaded footage the same way.

Can it translate the transcript? Yes. Vizard Agent translates from the transcript rather than re-listening.

Does it keep the timings? Yes. Word-level timings sit behind both outputs.

Can I get one document for a whole series? Yes. Vizard Agent orders them as chapters into a single manuscript.

Will it invent words it could not hear? It marks uncertainty rather than guessing confidently. Check specialist terms yourself.

Can it write a blog post from it? Yes, but say so. A cleaned transcript and an article are different documents.

Does cleaning ever cut too much? It can, which is why Vizard Agent checks the chapter lengths afterwards.

Can it caption the clips with this text? Yes. The subtitle file it produced is the one it burns in.

Why two files instead of one? Because captions break on breath and prose breaks on meaning. One file cannot be good at both.

Can it find a quote for me? Yes. Ask for the moment and Vizard Agent searches the transcript and gives you the timecode.

Does it work on a very long recording? Yes. Vizard Agent transcribes in passes and joins them rather than truncating the tail.