How to iterate on a design by changing one thing at a time
Tell Vizard Agent to produce still mockups rather than video, keep one version as the untouched baseline, and change a single variable in each variation. It assembles them onto one comparison board with the change written under each panel, and confirms that nothing except the named variable moved.
What is the short version?
You are deciding a layout — a thumbnail, a header, a caption position — and each round of "a bit bigger, maybe a different colour" costs a render and settles nothing. Stills cost almost nothing and they answer the question. Vizard Agent works that way if you ask.
- Tell Vizard Agent to produce stills only and not to touch the video.
- Name one version as the baseline and vary one thing at a time.
- Ask it to confirm only that one thing changed.
What do you need before you start?
The material and the decision you are actually trying to make. "Which of these reads better at thumbnail size" is a question stills can answer; "does this feel premium" is not, and no number of variations will settle it.
Have the specifics to hand if you have them — a hex value, a font weight, a pixel inset. Those make a variation reproducible, and they are what turns "warmer" into something that can be applied and then checked.
What do you type into Vizard Agent?
Tell Vizard Agent stills only, and say it explicitly. Otherwise a request to try a different colour quite reasonably reads as a request for a whole new version of the video, which is by far the more expensive of the two interpretations available.
Do not edit any video. Produce still mockups only — two thumbnails at 1280×720 on this background.
Then, on the next round:
Keep Mockup 1 as the baseline and produce three variations, each changing only one thing.
That last clause is the whole method. Three variations that each change three things tell you which one you like and nothing about why.
What does Vizard Agent actually do?
It builds the stills, puts them on one board, and writes the change under each panel. The notes matter as much as the images — a week later a board of four thumbnails with no annotations is unusable, and even an hour later it is hard to remember which was which.
In that session the passes ran:
- Built four cover mockups from the video's own frames so they could be compared side by side.
- Delivered them as one comparison sheet plus the individual files.
- Produced still-only comparisons with the font, size, colour, position and highlight details written beneath each frame.
- Confirmed each time that the single PNG was the only output and no video had been exported.
- Built a four-way board from an unchanged baseline with a change note under each variation.
- Checked the two mockups together to confirm that only the newly added wordmark had changed.
That last check is the one that makes iteration trustworthy. Drift is the characteristic failure of this process — a variation that also moved the margin by four pixels is not a test of the thing you were testing.
What does the result look like?
One image per round, with the variations labelled and their differences written down underneath them. You reply to Vizard Agent with a letter rather than a paragraph, and the next round starts from that letter instead of from a re-description.
Matching happens by measurement rather than by eye. When a new wordmark had to sit with an existing label, the label's real settings were read off — a specific weight, a specific size, a specific letter spacing — and applied, then the alignment was confirmed in both stills rather than assumed.
When does this not work well?
When the thing you are deciding only exists in motion. Timing, transitions and the way a caption enters the frame cannot be judged from a still image at all, and a board of frames will actively mislead you about all three of them.
It also stops helping somewhere past about four variations. Beyond that nobody can hold all the differences in mind at once, and the answers start coming out arbitrary — two rounds of four from Vizard Agent beats one round of twelve every time.
And a question that is really about taste will not resolve. If you cannot say what would make one option win, more variations are a way of postponing the decision rather than making it.
How do you fix a result that came back wrong?
If a variation changed more than it was supposed to, say which extra thing moved. Vizard Agent rebuilds that single panel from the baseline rather than redoing the whole board, which keeps the rest of the comparison valid.
If none of them work, change the variable rather than the amount. Four sizes of the same wrong element is a wasted round; the useful next step is usually a different element entirely.
If you cannot tell them apart, the variation is too small to matter. Ask for a bigger step — and if the bigger step also makes no difference, that is a real answer about where to spend your attention.
How does Vizard Agent compare to doing it yourself?
In a design tool this is what duplicating a layer and renaming it is for, and it works well right up to the point where you have nine artboards and no notes. The discipline is not technical; it is remembering to change one thing and write down what it was.
What helps is the board and the drift check. Having the variations and their change notes in a single image makes the comparison honest, and verifying that nothing else moved is the step that stops a promising result being an accident.
Common questions
Will it export a video by mistake? Not if you say stills only. Vizard Agent confirms nothing was rendered.
How many variations at once? Three or four. Beyond that Vizard Agent will tell you the comparison stops being useful.
Should the baseline change? No. That is what makes it a baseline, and Vizard Agent leaves it untouched.
Can I get the specs written on the board? Yes, under each panel. Ask Vizard Agent for the font, size, colour and position.
How does it match an existing element? By reading its actual settings rather than approximating them by eye.
Can I give exact values? Yes, and it helps. Vizard Agent applies a hex or a pixel inset exactly.
What if I want to vary two things? That takes two variations from Vizard Agent. One each, or you learn nothing from either.
Does this work for captions? Yes, for size and position. Vizard Agent cannot show you how they animate in a still.
Can it save the winner? Yes. Ask Vizard Agent to store the approved settings as a named preset.
How do I change a saved preset? By value — a shift in pixels, a new inset. Vizard Agent updates the file and re-delivers it.
Are the stills from my own footage? They can be. Vizard Agent pulls frames from the video so the mockups are real.
Can I get the individual files too? Yes. Vizard Agent delivers them alongside the board.
Does a board cost less than a render? Considerably less, which is most of the reason to work this way.
What if the difference only shows at small size? Ask Vizard Agent to include a thumbnail-sized version of each panel.