How to get ten versions of the same video to choose between
Tell Vizard Agent you want one video in ten treatments, not ten videos. It builds the shared parts once — the narration, the subtitle timings, the cut — then varies only the axes you name, renders the set in parallel, and puts frames from all ten on a single contact sheet so you can compare them without watching ten files.
What is the short version?
"Make me ten versions" is ambiguous, and the expensive reading of it is the wrong one. Ten separate edits cannot be compared against each other in any useful way, because every difference between any two of them is a different kind of difference, and you end up choosing on a hunch.
- Tell Vizard Agent the edit is fixed and the treatment varies.
- Name the axes: voice, music, pace.
- Ask for one contact sheet covering all of them.
What do you need before you start?
The footage or the finished cut, and a decision about what you are actually testing. Vizard Agent can vary almost anything, but a set where everything varies at once is not a test — it is ten videos, and you will pick one on a hunch.
- The cut, or the footage. The thing being treated.
- The axes. What is allowed to differ.
- The count. Ten is manageable; fifty is not.
- Where they are going. It sets the frame and length.
- Whether you want all of them. Usually yes, to compare.
What do you type into Vizard Agent?
Say "the same video". Those three words do most of the work here, because the default reading of "ten videos" is ten separate pieces of work rather than one piece of work in ten different coats of paint — and Vizard Agent will happily do the expensive thing if that is what you appear to have asked for.
Prompt
Variants worth knowing:
- "The same video ten times over." Removes the ambiguity.
- "Same script, same shots." The constants, named.
- "Give me every download link." You are choosing, not approving.
What does Vizard Agent actually do?
Here is the Vizard Agent sequence on a real set of ten gameplay variants. Notice that almost all of the work happens once, before any individual variant exists at all, which is exactly what makes a set of ten cost barely more than a set of two.
- Finds the meaningful scene boundaries in the source.
- Auditions voices — an energetic one, a low one, a playful one.
- Generates each line of the narration.
- Generates three music beds in different moods.
- Assembles one shared speech track for all ten variants.
- Takes exact timings from that track for the subtitles.
- Builds one subtitle treatment used by every version.
- Measures the level of each music bed before the batch mix.
- Renders all ten variants in parallel.
- Builds one contact sheet of all ten and collects the links.
Step five is what makes the set comparable. One narration recorded once means the only audible difference between two variants is the thing you asked to vary, rather than a take that happened to land better.
Step eight prevents the laziest kind of false result. Three music beds at three different loudnesses will make you prefer the loud one, so Vizard Agent measures and matches them before the batch runs.
Step ten replaces ten playbacks with one page. Frames from the start and the middle of every variant, side by side, tell you which three are worth watching properly — and the other seven you never open.
What does the result look like?
Ten files that differ in exactly the ways you asked for and in no other way, plus one sheet showing all of them at a glance. Vizard Agent hands over every download link rather than a recommendation, because the choosing is the point of the exercise and the comparison only means something if the set is otherwise identical.
When does this not work well?
A variant set is for choosing between treatments, not for finding out whether the edit works. If the cut itself is wrong, all ten will be wrong, and the sheet will show you ten versions of the same problem.
- An unfinished cut. Fix the edit before multiplying it.
- Too many axes at once. Nothing is attributable.
- Very long videos. Ten renders of ten minutes is a lot of waiting.
- Genuinely different creative. That is separate work, not a variant.
- More than about a dozen. You will stop looking properly.
How do you fix a result that came back wrong?
Talk about the axis, not the file. Because the shared parts were built once, an instruction like "all the voices are too slow" changes the set rather than one member of it, and Vizard Agent re-renders the whole batch from the same base.
- "They are all the same." The axes are widened.
- "This is ten different edits." Rebuilt from one shared cut.
- "The music drowns the voice." The beds are re-measured and re-mixed.
- "I only want the three arcade ones." The set is narrowed and re-rendered.
How does Vizard Agent compare to doing it yourself?
By hand you make one, duplicate the project, change the music, export, and repeat. By version six you have lost track of which is which, two of them have a stale subtitle track, and the export queue has been running all afternoon.
| By hand | Vizard Agent | |
|---|---|---|
| The narration | Re-recorded or copied per version | Built once, shared |
| Music levels | Whatever each track came at | Measured and matched |
| Rendering | One after another | In parallel |
| Comparing | Ten playbacks | One contact sheet |
| Knowing which is which | A naming convention you forget | Labelled by its axis |
Common questions
How many versions should I ask for? Ten is a good ceiling. Vizard Agent will make more, but you will stop looking.
Can I vary just one thing? Yes, and that gives the cleanest comparison. Tell Vizard Agent which one.
Will the subtitles match across versions? Yes. Vizard Agent times them once from the shared narration.
Can the pace differ? Yes. Vizard Agent can speed a variant up without re-timing the subtitles by hand.
Do I get all of them? Yes. Vizard Agent uploads every variant and gives you the links.
Does it cost ten times as much? No. Most of the work is shared, and Vizard Agent only repeats the render.
Can I see them without downloading? Yes. The contact sheet Vizard Agent builds shows frames from every variant.
What if I want different footage per version? Then it is not a variant set. Ask Vizard Agent for separate edits instead.
Can it label which is which? Yes. Vizard Agent names each variant by the axis that makes it different.
What if two sound identical? Tell Vizard Agent and it widens that axis rather than adding more versions.
Can I test thumbnails this way? Yes, the same approach works for frames and end cards.
Should the music be the same loudness? Yes, or you are testing loudness. Vizard Agent measures the beds first.
What if I like parts of two versions? Say which parts. Vizard Agent can build a combined version from the same base.
Can I come back and add more later? Yes. The shared narration and subtitles are reused, so extra variants are cheap.