PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts

daniel
daniel

Posted on

I Accidentally Turned a Small Market Into a Festival

I started with a pretty ordinary event brief.

Saturday morning. A small outdoor market with coffee, baked goods, handmade ceramics, and a musician near the stalls.

That was enough for the idea.

Then I started writing the video prompt.

Somewhere along the way, the small market picked up a sweeping entrance shot, a much bigger crowd, rows of busy stalls, golden light, extra performers, and a dramatic final view.

It sounded good when I read it back.

It just wasn't the same event anymore.

Back to the Original Notes

I deleted most of what I'd added and looked at the brief again.

Coffee being poured made sense.

Someone browsing the ceramics made sense.

The musician could stay in the background.

One wider view could show the outdoor setting.

I didn't need much more than that.


I kept the same notes open when I made a few Wan 3.0 drafts. That made it easier to spot the moment a scene started drifting.

In one version, the market looked much larger and busier than I'd imagined it. There was nothing obviously wrong with the image itself. If I'd seen it without the brief, I probably wouldn't have questioned it.

But it wasn't what I'd started with.

So I went back rather than trying to make the new version fit.

Some Things Can Change. Others Can't.

A fictional pottery workshop gave me a similar problem.

My notes said:

Sunday at 2 PM

Beginner class

Materials included

Eight places available

There's plenty I can invent visually around that.

The tables can be wooden. There can be clay on the workbench. Finished cups might sit on shelves in the background.

But I can't quietly turn Sunday into Saturday.

And if there are eight places, I shouldn't turn that into twenty because a fuller room happens to look better.

The same goes for a real price, address, product name, opening time, or offer.

Those details come from the source, not from whatever looks nicest in the generated scene.

This matters more than I expected because once a draft looks polished, invented details stop feeling invented.

They just look like part of the video.

I Don't Need to Show the Whole Brief

I used to make this harder by trying to turn every line into a shot.

Say a bakery brief mentions fresh bread, takeaway coffee, outdoor seating, opening hours, pastries, and a weekend menu.

That's useful information.

It doesn't all need screen time.

Maybe the video only shows the first part of the morning: the door opens, bread comes out, coffee is poured, and someone leaves with a paper bag.

The menu can stay on the page.

The opening hours can stay in the caption.

That leaves the video with one simple job instead of six.

Sometimes the bigger problem is that the original notes don't give me anything useful to show.

"Great food, friendly atmosphere, perfect weekend spot" sounds fine in copy, but I can't do much with it visually.

"Small bakery counter, cinnamon rolls coming out of the oven, two tables outside, customers picking up coffee" is much easier.

It's not better because it sounds more creative.

It's better because I can see what happens.

I Check the Draft Against the Thing I Started With

When a draft is ready, I still keep the original notes nearby.

I'm looking for obvious visual problems, but I'm also checking for smaller changes I might have stopped noticing.

Maybe the shop has become much bigger.

Maybe there's a service in the scene that the business doesn't offer.

Maybe a quiet local event suddenly looks packed.

Maybe a product appears to do something that was never part of the source material.

With a fictional exercise, I can decide whether any of that matters.

With a real business, event, or product, I don't make that assumption.

The same goes for actions that look convincing on screen. If generated footage appears to show how food is prepared, how a product works, or how a tool is used, I don't treat the generated motion as an instruction or proof that the action is accurate. I go back to the real source before using that kind of detail publicly.

A polished draft can still get the basic facts wrong.

That's why the boring notes stay open on my screen.

I started with a small Saturday market.

I'd like it to still be a small Saturday market when I'm finished.

Top comments (0)