I had written a voiceover called “Here’s To Dreaming.” It was one of those reflective pieces that arrives late at night when the house is quiet and the ideas are louder than the room.
The piece was about possibility.
About making the invisible visible.
About the strange experience of spending a lifetime building things and still waiting for the next person who wants to come out and play.
I needed visuals.
Naturally, I assumed that would be the easy part.
I was wrong.
2/9
My first instinct was exactly what you would expect.
Dream imagery.
Light.
Particles.
Ideas becoming visible.
The usual AI-assisted visual poetry.
The problem was that it felt generic almost immediately.
The images were pretty.
They were also boring.
There is a peculiar danger when working with AI image generators. They are astonishingly good at producing things that look impressive while saying absolutely nothing.
I kept generating images.
The images kept looking like everyone else’s images.
That’s when I realized I wasn’t trying to illustrate creativity.
I was trying to illustrate conflict.
3/9
The breakthrough came from a simple question:
What if the battle inside the mind was an actual battle?
Not symbolic.
Not abstract.
Not diagrams floating through space.
An honest-to-goodness battlefield.
Armies.
Smoke.
Movement.
Pressure.
Conflict.
The voiceover was calm.
The imagery could be chaos.
The juxtaposition immediately felt more interesting.
Instead of visual agreement, I suddenly had visual tension.
The voice said reflection.
The images said struggle.
The combination was much stronger than either one by itself.
Lab note:
When something feels cliché, don’t polish it.
Change the premise.
4/9
The next problem was historical setting.
Future battle?
Too much science fiction baggage.
Modern battle?
Too much real-world baggage.
Historical battle?
Now we were getting somewhere.
I eventually settled on a Revolutionary-era battlefield.
The appeal wasn’t historical accuracy.
The appeal was distance.
A watercolor battle from another century feels like memory rather than journalism.
That distinction matters.
I wasn’t making a war film.
I was making a dream.
5/9
Then Midjourney taught me another lesson.
The prompts kept saying “watercolor.”
The images kept saying “Hollywood.”
I would ask for watercolor battlefields.
Midjourney would give me a cinematic war movie with a few artistic flourishes.
At first I assumed the model wasn’t listening.
The model was listening.
It was obeying my strongest instructions.
The problem was me.
I was telling it:
watercolor
and
cinematic realism
at the same time.
Guess which one won?
Lab note:
AI models don’t read prompts.
They resolve conflicts.
6/9
The fix turned out to be surprisingly simple.
Stop treating watercolor as a style.
Start treating watercolor as the medium.
Instead of:
“watercolor battlefield”
I switched to things like:
“watercolor painting on textured paper”
“ink wash bleeding edges”
“visible pigment diffusion”
“hand-painted illustration”
I also removed many of the realism cues.
No more obsessive lens descriptions.
No more cinematic still language.
No more realism-first instructions.
Almost instantly the images transformed.
Not because Midjourney got smarter.
Because I finally became clearer.
7/9
The funny part is that the same lesson appeared elsewhere in the project.
I was testing music for YouTube.
The voiceover worked.
The instrumental worked.
A vocal segment triggered copyright detection.
So began the usual round of experimentation.
Tempo changes.
Pitch changes.
More testing.
More uploading.
Eventually I discovered that a small key shift solved the problem entirely.
Not a giant change.
Just enough.
Creative work often looks like inspiration from a distance.
Up close it looks suspiciously like troubleshooting.
8/9
Tools & Creative Stack
Voiceover
- ElevenLabs
Images
- Midjourney
Editing
- Kdenlive
Music experimentation
- LALAL.AIStem separation and audio editing workflow
Writing
- ChatGPT
Budget
The cost wasn’t money.
The cost was iteration.
The project became another reminder that the biggest expense in AI-assisted creativity is often not generation.
It’s discernment.
Knowing what to keep.
Knowing what to throw away.
Knowing when the machine is wrong.
Knowing when you’re wrong.
9/9
The Real Lesson
The most important discovery wasn’t about Midjourney.
It wasn’t about prompts.
It wasn’t about copyright systems.
It was about conflict.
The project only became interesting when I stopped trying to illustrate the voiceover and started arguing with it.
A quiet reflective narration floating over watercolor battlefields shouldn’t work.
And yet it does.
Because creativity itself often feels exactly like that.
Part dream.
Part conflict.
Part construction.
Part chaos.
The invisible doesn’t become visible peacefully.
Usually it has to fight its way into existence.
TL;DR
I started with dream imagery.
I ended with watercolor warfare.
The AI wasn’t confused.
My prompts were.
The battle wasn’t in the video.
The battle was learning how to make the video.
Steve Teare
video alchemist
TerminallyBored.Monster
Palouse, Washington USA