How to make a compilation where nobody appears twice

Last updated 2026-09-04 · 7 min read

What is the short version?

A montage of people is not a montage of shots. Repeating a shot is a mistake; repeating a person looks like you ran out of guests, which is a much worse thing to accidentally say about your own show.

1. Go to Vizard Agent and describe the compilation.
2. Say no person may appear more than once.
3. Say where the rest of the episodes are, so it can go and get them.

What do you need before you start?

A way for it to reach more material than the compilation needs. The instruction "no repeats" is only useful if there is somewhere to go when the shortlist runs short — a channel, a playlist, a folder of episodes.

• The library. A channel, playlist or folder.
• The rule. No person twice.
• The order. Who closes it, if that matters.
• What counts as an appearance. On screen, or speaking.
• Name tags, if you want them. It can build them.

What do you type into Vizard Agent?

Say precisely what must not repeat. "No repeats" on its own is ambiguous in a way that matters here — a compilation can reuse an episode without reusing a person, and it is nearly always the person you mean rather than the source Vizard Agent took them from.

Prompt

Variants worth knowing:

• "Pull from episodes you haven't used." Permission to go and get more.
• "Keep X at the end." One fixed position, the rest free.
• "No repeats of the host either." Usually worth saying.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real podcast intro that kept repeating guests. The interesting part is that the fix was not a better selection from what it had — it was going to get more.

1. Finds the podcast's own channel and its prominent guests.
2. Browses the channel for the guest interview episodes.
3. Pulls the real episode links from the channel and the website.
4. Parses the full video list to find every guest episode.
5. Downloads several episodes at once and retries the failures.
6. Scans each episode to see how the guests appear on screen.
7. Reviews each episode's layout — the panes differ per episode.
8. Finds a lively on-camera moment for each guest.
9. Grabs frames at those moments and measures the video panes.
10. Searches for further episodes when more guests are needed.
11. Downloads eight more unused episodes in parallel.
12. Looks up each new guest's role for the name tags.
13. Builds brand-style name tags for every guest.
14. Locates a mid-episode layout switch that moved a guest's pane.
15. Renders the version with all-unique guests.
16. Verifies that no person repeats and that the edit is clean.

Step eleven is the behaviour worth knowing about. Asked not to repeat, the useful response is to widen the pool rather than to make a tighter selection from a pool that is too small — so it went and downloaded eight episodes it had not touched.

Steps six, seven and fourteen are the unglamorous half. In a multi-guest podcast the guest is in a pane, the pane moves between episodes, and it sometimes moves within an episode. Cropping to a fixed position gets you a shot of an empty chair.

What does the result look like?

A compilation where every face is a different person, each in a moment where they are actually doing something, with their name and role on screen — and the person you wanted at the end is at the end.

The uniqueness is verified rather than assumed. That matters because the failure is easy to miss: the same guest in two different shirts from two different episodes reads as two people until someone who knows the show watches it.

When does this not work well?

The rule needs a pool deep enough to satisfy it. Where there are simply fewer distinct people than there are slots to fill, no amount of searching by Vizard Agent fixes that, and the honest options are a shorter compilation or a relaxed rule.

• More slots than people. Something has to give.
• A closed library. Nothing to widen into.
• Guests who appear together. One clip, two people.
• Very similar-looking people. Say how to tell them apart.
• Episodes it cannot fetch. Some sources block downloads.

How do you fix a result that came back wrong?

Name who repeated or who is missing. Vizard Agent keeps the episode list, the per-episode layouts, the chosen moments and the name tags, so swapping one guest is a re-cut rather than a fresh trawl of the channel.

• "That is the same person twice." One instance replaced from an unused episode.
• "You cut to someone else first." The lead-in frames trimmed.
• "Wrong role on the tag." Looked up again and corrected.
• "Put X last." Reordered; uniqueness re-verified.

How does Vizard Agent compare to doing it yourself?

By hand the work is not the editing, it is the sourcing — finding which episodes you have not used, downloading them, and finding a two-second moment in each where the guest looks alive. That is an afternoon, and it is why compilations quietly repeat people.

Common questions

Will it really go and find more? Yes. Vizard Agent widens the pool rather than reusing what it has.

Can it use my own folder instead? Yes. Point Vizard Agent at the files and it works from those.

What if two guests share a clip? Vizard Agent tells you rather than counting it as one appearance.

Can I fix who goes last? Yes. Name them and Vizard Agent builds the rest around it.

Does it add name tags? Yes. Vizard Agent builds them in your style, with the roles researched.

What if it cannot download an episode? Vizard Agent retries, then tells you which ones it could not reach.

Does it check the host repeats too? If you tell Vizard Agent to. Worth asking for explicitly.

Can it tell two similar people apart? Usually. Give Vizard Agent a note if two guests look alike.

How many people can it handle? As many as the library has. Six to a dozen is typical for an intro.

What if a guest's pane moves mid-episode? Vizard Agent looks for the switch rather than cropping blind.

Can it pick lively moments specifically? Yes — that is what Vizard Agent looks for rather than taking the first shot.

Does it verify at the end? Yes. Vizard Agent checks uniqueness on the finished render.

Can it do this for products instead of people? Yes. The same rule applies to any repeated subject.

Why does repeating look so bad? Because a compilation is implicitly a claim about how much material you have, and a repeat quietly contradicts it.