Micro story

For the past few weeks, everyone’s been claiming you can create emotional AI video with “one magic prompt.”

So, Reader I decided to run un experiment.

I wanted to build a three-shot emotional micro-story using AI video .. same woman, same vibe, same apartment, same emotional arc.

But there was a problem.

Veo3 can’t use reference images in Text-to-Video.. which means you lose character consistency the moment you move to the next scene.

And if you’ve ever tried multi-shot AI storytelling, you know:

Consistency is the entire game.

So instead of throwing prompts on the wall hoping they stick, I built myself a little production pipeline, shall we call it.

Today, I wanted to share with you exactly what I did it.

I't dumb simple, really.

1. Create the character in Midjourney.

Generate a mid-30s Scandinavian woman until you find a face you want to use.

2. Then I Generated matching angles.

Use MidJourney to create additional angles of the same woman.
Pick the ones that best match her bone structure, eyes, and face shape.

3. I built a Character Sheet.

Assemble the selected angles into one reference sheet.

4. Do the same for the environments.

Generate a living room and bathroom, eg, in Midjourney.
Build a Scene Sheet for each location similar to the character sheet.

5. Go to Veo3 and enter Reference-to-Video mode.

Upload the character reference + environment reference.

6. Write a simple action-only prompt.

Do not describe:

appearance
facial features
outfit
environment details

The references handle that.
Your text should only describe what happens.

7. Generate Shot 1.

Run the prompt a few times until you get a result you like.
Download it.

8. Repeat for Shot 2.

Use the same character reference sheet.
Swap environment sheet.

Ok, someone on LinkedIn called out that I F****D up the aspect ratio... that would be this one..

Only change the action in the prompt.

9. Repeat for Shot 3.

Same references.
Only new action prompt.

10. Download all final clips.

11. Assemble in Premiere Pro.

Place each shot in sequence and adjust timing.

12. Upscale the footage.

Since Veo3 reference mode outputs 720p, upscale using Topaz Video AI.

13. Color grade.

Apply a bit of color grade after the upscale for best quality.

14. Final render.

15. Core rule to remember:

When using image references, only prompt actions.
If you describe appearances or environments in the text, Veo3 hates you.

One final *gassed up* takeaway

Had to get a little Hollywood with it (love the loft style toilette, haha)

Look how CRAZY close her sweater is in every shot.

I think that's absolutely insane!

But on the real,

If you’re a creative director, small-agency lead, or brand marketer:

AI won’t automatically solve all your problems.

The future isn’t a one-prompt fix all for creative and filmmaking.

While it's faster, its also slow sometimes, a little tedious. I find there is a lot of prompting, redundancy, waiting, errors, re-prompting, waiting.

Not to mentioned the clean up that's needed in post.

But my final honest verdict thus far?

At the same time VEO3 and AI video gen is fun as sh*t.

Great for short format content and video.

So far we're just testing basics behind the scenes, over here.

And my newsletter, haha.

But I see it becoming part of our offering soon.

Clients keep asking for it.

So we have to make sure we're on it.

Appreciate you being here in the middle of testing all this AI and animation stuff with me.

I'll speak to you shortly.

—Christian