How to fix a new episode that drifted from your earlier ones

Last updated 2026-09-05 · 8 min read

What is the short version?

A series drifts one small decision at a time — a slightly smaller title, a slightly different blue, a voice that sits a shade flatter. Nobody notices in isolation, and everybody notices when two episodes play back to back. The fix is a reference, not a description.

1. Go to Vizard Agent with the new episode.
2. Name an earlier one that looks and sounds right.
3. Say which parts of the new one drifted.

What do you need before you start?

An episode you are happy with and a sentence about what feels different. You do not need to know the font, the size or the colour — those are measured off the reference — but naming which episode is the standard matters more than describing what it looks like.

• The new episode. As delivered.
• An earlier one you consider correct.
• What drifted. Title, captions, voice, grade.
• What must not change. Usually the script and the cut.
• Any second reference, if two episodes differ.

What do you type into Vizard Agent?

Name the episode instead of describing the style. "The font was larger, slightly blue and more central" is useful, but "make it like the opening of that other video" is precise, because the other video is a measurable object and your memory of it is not.

Prompt

Variants worth knowing:

• "Like that one." A reference beats a description.
• "Keep everything else." Protects what did not drift.
• "It sounds more AI than the others." A voice complaint, not a script one.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real series of vertical food reels whose latest episode had drifted from the previous two. Almost every step is a measurement taken off the earlier videos rather than a decision made fresh.

1. Crops the title and caption areas out of the earlier episodes.
2. Inspects the title typography and the caption style close up.
3. Checks two earlier episodes rather than relying on one.
4. Measures the title wash timing and its colour.
5. Measures the caption and title geometry from the reference.
6. Checks how an earlier episode framed the same kind of shot.
7. Matches the reference title's font size and sizes the subline.
8. Builds the new title and compares it against the reference directly.
9. Identifies the caption font and compares it to the earlier ones.
10. Compares the colour grades between episodes.
11. Clones the narrator from the earlier episodes' audio.
12. Builds a voice comparison file and listens to the candidates.
13. Re-clones from a cleaner sample when the voice sounds synthetic.
14. Verifies the title and the voice on the finished render.

Step thirteen is the one worth remembering. "It sounds more artificial than the other videos" is usually a problem with the sample the voice was built from rather than with the voice settings, so the repair is a cleaner reference recording rather than a different dial.

Step three is a small habit that prevents an argument. Two earlier episodes may not agree with each other either, and finding that out before rebuilding anything is better than discovering it after you have matched the wrong one.

What does the result look like?

An episode that sits alongside the others without standing out: the title at the same size, in the same place, in the same colour; the captions in the same face; the narrator sounding like the narrator; the grade in the same family.

Played back to back, nothing tells you which one was made later.

When does this not work well?

Matching needs something to match against. If the earlier episodes were made elsewhere and you only have the exported files, most things can still be measured — but a font that is not available cannot be reproduced exactly, only approached.

• A font you do not have. It can be matched, not reproduced.
• Earlier episodes that disagree. Say which is the standard.
• A voice with no clean sample. Cloning needs clean audio.
• A deliberate change. Sometimes drift was somebody's decision.
• Heavily compressed references. Colour measurements get soft.

How do you fix a result that came back wrong?

Say which element still looks different, and say which earlier episode you are comparing it against. Vizard Agent keeps the crops and the measurements it took from that reference, so a second pass compares against exactly the same evidence rather than starting the whole comparison over again.

• "Still smaller than the others." Re-measured and resized.
• "The blue is off." Colour sampled again from the reference.
• "The voice is still robotic." Re-cloned from a cleaner sample.
• "The captions are a different face." Font identified and matched.

How does Vizard Agent compare to doing it yourself?

By hand this means opening two projects side by side, which is only possible if you still have them. Working from the exported episodes instead — cropping the title area and measuring it — is the approach that works whoever made the earlier ones.

Common questions

Do I need the old project files? No. Vizard Agent measures from the exported episodes.

Can it match a font I do not have? It gets close and tells you. Vizard Agent will name what it used.

What if two episodes look different? Say which is right. Vizard Agent checks both first.

Can it clone my narrator? Yes, from clean audio in the earlier episodes.

Why does the voice sound artificial? Usually the sample. Vizard Agent re-clones from a cleaner one.

Will it change my script? No. Vizard Agent matches presentation, not content.

Can it match the colour grade? Yes. Vizard Agent compares grades between episodes.

What about the music? Same approach — Vizard Agent analyses the earlier bed.

Can it check every element at once? Yes. Ask Vizard Agent for a full comparison against the reference.

Will future episodes stay matched? More reliably if you also hand Vizard Agent the rules in writing once.

Can it show me the two titles together? Yes. Vizard Agent builds a direct comparison.

Does it work across formats? Yes, though Vizard Agent has to re-measure the geometry per format.

What if I want to change the style now? Then say so; Vizard Agent will set the new standard instead.

Why measure rather than describe? Because "slightly bigger and a bit more blue" is a memory, and the earlier episode is a file that can be measured.