Vizard Agent

How to make a short story video about a renovation from nothing but a description

Last updated 2026-10-10 · 2 min read

Describe the story to Vizard Agent (a man clears the ground around an old house, then starts the work) and it builds the whole thing: a reference sheet so the man and the house stay the same, five generated scenes, narration, music and subtitles checked against the fonts actually installed.

What do you type into Vizard Agent?

Write the story in plain sentences in your own language. Vizard Agent works out the scenes, the characters and the shot order from that, so a couple of lines describing what happens beats a formal brief or a shot list. What it can't invent for you is the order of events, and that order is the whole appeal of this format.

Prompt

Make a video showing how a man renovates an old house. He should first clean up around it and then start the work. Around [40] seconds, vertical.

What does Vizard Agent actually do?

A process story needs every shot to show the same house and the same person, so Vizard Agent anchors both to one reference sheet. It also doesn't accept its own output on trust. A subtitle that is present but hard to read is the most common way a good video is spoiled, so it looks at the render and goes back. In the session behind this article it:

When does this not work well?

Everything here is generated, so the story is only as convincing as generated video allows. Hands doing physical work are where it strains most, and the house can still drift between scenes despite the reference sheet. It isn't your house: for a real renovation, your own photographs make a better and more honest video, and a generated before-and-after is fiction.

Common questions

Do I need any footage? No. Vizard Agent generates every scene from your description, and your own photographs make it better if you have them.

How many scenes fit in forty seconds? About five. Each holds for roughly eight seconds, which suits a process story.

Will the subtitles be readable? Vizard Agent checks the rendered result rather than trusting the file, and you can ask for larger text.