How to edit a recorded lecture to a written spec

Last updated 2026-08-16 · 5 min read

What is the short version?

Lecture editing is production work with a written standard behind it: platforms, aspect ratios, caption rules, an intro and outro that never change. The creative decisions were made once, months ago. What remains is applying them consistently to every recording, which is the least interesting part of the job and the part that takes the day.

1. Go to Vizard Agent and upload every clip from the session.
2. Paste your specification document as the prompt.
3. Keep the series in one project so each lecture matches the last.

What do you need before you start?

The raw takes and your written spec. Vizard Agent does better with a specification pasted verbatim than with a summary of it, because the details you would trim as boilerplate — platform, audience, caption placement — are exactly the constraints that decide the edit.

• Every clip, including the retakes. Vizard Agent transcribes all of them and works out which take to use.
• The specification as written. Platforms, target audience, format, resolution, caption rules, anything about the intro and outro.
• The previous lecture, if there is one. Same project, or upload it. This is what keeps a series looking like a series.
• Anything the spec leaves implicit. House rules that live in your head rather than the document are the ones that get missed.
• A note about slides or screen recordings, if they arrive as separate files.

What do you type into Vizard Agent?

Paste the whole document. It reads as too long for a prompt and it is not: every line of a spec is a constraint, and Vizard Agent applies constraints it can read far more reliably than ones you expect it to infer from a sentence of description.

Prompt

Variants worth knowing:

• Same as last time. Once one lecture is approved, "same spec as the previous lecture" is the whole prompt.
• Multi-platform. Say which platforms and Vizard Agent delivers a version per destination rather than one compromise.
• Fix a spoken mistake. You can ask for a specific word to be replaced in the audio, which is a common request on lecture material.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real five-clip lecture. It is a long run, and the shape of it is worth seeing: everything about the previous delivery and every transcript is gathered before a single edit decision is made.

1. Reads the project's own record of what has been delivered before.
2. Probes all five source clips for their technical properties.
3. Downloads the clips and the previous delivered lecture.
4. Loads its talking-head guidance rather than improvising an approach.
5. Samples frames from every clip and from the previous final, then looks at them.
6. Checks the transcription and caption tool options before transcribing.
7. Transcribes all five clips.
8. Inspects the previous delivery's look so this one matches it.
9. Checks the rotation on a sideways clip and compares rotated versions to find the right orientation.

Step 8 is why a series holds together, and step 9 is the kind of thing that quietly ruins a lecture edit: one clip recorded on a phone held the other way, discovered at export time rather than at the start.

What does the result look like?

Delivery follows the specification you pasted rather than the properties of the source files, which is the whole point of working this way: resolution, aspect ratio, caption placement, and the intro and outro all come from your document, and Vizard Agent treats each line of it as a constraint to satisfy rather than context to consider.

This category was not separately measured for timing. Comparable work runs a median of 28 to 38 minutes end to end, and across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits. A five-clip lecture sits at the longer end, because every clip is transcribed and inspected before anything is cut.

When does this not work well?

The failure mode here is a spec that is not really a spec. Vizard Agent applies what is written; where the document is silent, it makes a reasonable choice, and reasonable is not the same as your house style.

• Unwritten house rules get missed. Anything your editor "just knows" is invisible. Write it into the spec once and it stops being a problem.
• Audio problems need naming. Room echo, a noisy air conditioner, inconsistent levels between takes. Vizard Agent can address these, and will not necessarily know they bother you.
• Sideways phone clips. Rotation is stored separately from the picture, so a clip can arrive lying down. Vizard Agent checks for this, and if one still comes back wrong, saying so fixes it.
• Very long recordings cost accordingly. Transcription and inspection scale with duration, so a two-hour lecture is a different job from a twenty-minute one.
• Correcting what was said is limited. Replacing a word in the audio works. Rewriting a paragraph of a lecture does not, unless you are willing to have it re-voiced.

How do you fix a result that came back wrong?

Quote the line of the spec that was not followed, or point at the timestamp. Vizard Agent keeps the transcripts, the frame samples and the previous delivery reference, so a correction is applied to the existing edit rather than sending it back through the whole analysis.

• "The captions should be at the bottom, per the spec." Applied without re-transcribing.
• "Replace the word 'four' with 'three'." A targeted audio fix, as requested in the run behind this page.
• "Use the second take of that section." Vizard Agent already transcribed all of them.

How does Vizard Agent compare to doing it yourself?

An editor working to a spec is doing skilled work at low creative stakes: they know what to do, and doing it takes a day per lecture. The value is in the consistency, and consistency is precisely what is expensive to maintain across a long series.

Common questions

Do I have to paste the whole specification? It is better if you do. Vizard Agent applies written constraints more reliably than described ones.

Can it pick between multiple takes? Yes. It transcribes every clip, so it can compare what was said as well as how it looked.

Can it deliver different versions per platform? Yes. Name the platforms and Vizard Agent delivers one per destination.

What about slides or screen recordings? Upload them with the takes. Vizard Agent treats them as additional source material and will cut to them where the transcript says you are referring to what is on screen.

Will lecture forty look like lecture one? If they are in the same project, yes. Vizard Agent checks the previous delivery each time rather than relying on remembered settings.