How to revise a video by sending notes at timecodes

Last updated 2026-09-04 · 7 min read

What is the short version?

This is how people actually review a video — not with a document of consolidated feedback, but by watching it, stopping, saying one thing, and watching again. Vizard Agent works that way round too, so you do not have to save your notes up.

1. Go to Vizard Agent and watch the current cut.
2. Send one note with the timecode and what to change.
3. Send the next one when you have watched the result.

What do you need before you start?

A cut to react to, and nothing else. You do not need to know the right wording before you ask — most of the notes in a real revision round are "make that smoother" rather than a replacement sentence, and that is a perfectly workable instruction.

• The current version. Whatever it is.
• The timecode. Roughly is fine.
• What is wrong. In your own words.
• Or the replacement text. If you have written it.
• Nothing consolidated. One note at a time is the point.

What do you type into Vizard Agent?

Say where and say what. You do not have to be formal about the timecode — "1.02" and "at 1:02" and "the bit about the piano" all land, because Vizard Agent is holding the transcript and can find the place from the words as easily as from the number.

Prompt

Variants worth knowing:

• "Smooth that please." No replacement text needed.
• "Actually put it back." Reversals are expected.
• "Cut the first sentence of that paragraph." Position, not timecode.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real personal-story video that went through twenty-seven separate notes. Steps seven and eight of the revision run are the ones worth seeing: a reversal, and a repair the note did not ask for.

1. Downloads the source recording and checks the file.
2. Transcribes it and looks at frames from the video.
3. Builds a readable transcript and tidies the punctuation artefacts.
4. Polishes the text and uploads it for you to read.
5. Removes the opening paragraph — note one.
6. Cuts the aside about the guests arriving — note two.
7. Smooths the paragraph at 1:02, then drops its last sentence.
8. Puts that sentence back when you decide you wanted it.
9. Replaces the paragraph at 1:47 with your new wording.
10. Takes out a sentence at 2:13 — then fixes the word left dangling by it.
11. Rewrites the 2:13 paragraph with the text you sent.
12. Works through 3:00, 3:22, 3:57, 5:37 and 6:24 as each note arrives.
13. Removes a duplicated sign-off it spotted while working.
14. Builds a document version of the finished text.
15. Reads the word timings and finds the exact story start.
16. Generates the subtitles and renders a short test with them.
17. Looks at the captions on a real frame.
18. Strips the filler words from the captions.
19. Renders the trimmed video, checks the frames, and delivers.

Step ten is the detail that separates this from search-and-delete. Removing a sentence usually leaves something behind — a conjunction, a "so", a "and then" that was pointing at the sentence you just cut. Vizard Agent reads the join afterwards and repairs it, because a clean cut in a transcript is rarely a clean cut in the reading.

Step eight matters for a different reason. Nothing is destroyed as you go, so changing your mind three notes later costs one sentence, not a rebuild.

What does the result look like?

A finished video whose script has been through twenty-seven surgical changes, each one only touching what it named. Alongside it, a document version of the final text, and captions generated from the final wording rather than from the original recording.

The last part is what makes the method safe. Captions written early and patched afterwards drift out of step with the edits; here the subtitles come after the text is settled.

When does this not work well?

Note-by-note revision is a conversation, and like any conversation it can lose the thread. It is at its worst when the notes stop being local — when what you actually want is a different video rather than a corrected one.

• A structural rethink. Say that, rather than sending forty notes.
• Notes that contradict each other. It will ask which you meant.
• Vague timecodes on a long video. Quote the words instead.
• Batches of twenty at once. Fine, but harder to reverse cleanly.
• Changes to the picture, not the words. Say what you see, not the time.

How do you fix a result that came back wrong?

Send another note. That is the whole model, and it is why nothing here needs a formal correction process: Vizard Agent holds the full transcript, the word timings and every previous change, so correcting an over-eager edit costs one more sentence rather than a restart from the original recording.

• "That cut too much." Restored from the held transcript.
• "Wrong 2:13 — I meant the next one." Re-applied at the right place.
• "Undo the last three." Rolled back in order.
• "Now regenerate the captions." Rebuilt from the current text.

How does Vizard Agent compare to doing it yourself?

By hand this is the loop that eats an afternoon: open the project, find 2:13, make the change, re-render, watch, find the next one. The changes are trivial. The overhead around each one is not, which is why feedback usually arrives in one intimidating list instead.

Common questions

Do I need exact timecodes? No. Vizard Agent finds the place from the words if you quote them.

Can I send several notes at once? Yes. Vizard Agent works through them in order.

What if I change my mind? Say so. Vizard Agent puts back what it removed.

Will it edit anything I did not mention? No, beyond repairing what a cut left dangling — and Vizard Agent says when it does.

Can I edit the text myself instead? Yes. Ask for the document, edit it, and send it back.

Does each note re-render the whole video? The text changes are cheap. Vizard Agent renders when the script settles.

Can it show me the change before rendering? Yes. Ask for the revised paragraph as text first.

What about the captions? Generated at the end, from the final wording, so they always match.

Can it keep a record of the changes? Yes. Vizard Agent can deliver the script as a document at any point.

How many rounds is too many? Vizard Agent sets no limit. Twenty-seven is a real number from one project.

Can I change the voice at the same time? Yes, but say so separately — it is a different kind of note.

Does it get slower as notes pile up? No. Vizard Agent works from the current text, not the history.

Can someone else send notes too? Yes. Vizard Agent takes them in the order they arrive.

Why not send all my feedback at once? You can. But half of real notes only occur to you after seeing the previous one applied.