There’s a specific kind of frustration that feels like staring at a door that almost opens.
Not locked. Not broken. Just… uninterested in you.
That’s what this project became.
Two factory workers. Early 20s. Industrial dystopia. Cold steel. Wet concrete. Fluorescent light that makes everyone look like they’ve been slightly erased by the world.
And underneath it: a simple idea I could not get the machine to hold onto.
Recognition.
Not romance. Not fantasy. Not transformation.
Just the moment two people realize they are not alone in being different.
2/11
It started clean enough.
A 12-shot sequence. A controlled emotional arc. Midjourney for stills. Grok for motion. A straight pipeline.
Girl. Boy. Factory shift life. Distance. Lunch breaks. Accidental proximity. A reveal. Collapse into relief. Walking home.
Simple.
Except nothing about it behaved simply once it touched the model.
The images kept flattening into something else.
Portraits.
Beautiful, yes. But wrong in a very specific way.
They kept becoming “cinematic couple photography” instead of “two workers slowly noticing each other in a hostile industrial world.”
Lab note: Midjourney does not understand narrative intention. It understands archetypes.
3/11
Then came the first structural issue.
They kept looking at the camera.
Every shot.
Even when I asked for distance, for candid framing, for observational geometry.
It didn’t matter.
The model kept pulling them forward into direct eye contact like it was trying to convert the whole story into a series of album covers.
At some point I realized what was happening:
The reference image wasn’t just guiding identity.
It was enforcing portrait behavior.
Lab note: identity anchors come with posture. Not just face.
4/11
The second issue was worse.
The elf ears.
This was the core narrative device. The entire emotional spine of the project.
Hidden sameness. Accidental exposure. Shame. Recognition. Relief.
But Midjourney did what it always does when something isn’t physically enforced:
It erased it.
Or softened it.
Or quietly “corrected” it back into human normality.
And I kept trying to fight it with language.
Subtle. Slight. Barely visible.
Which, in Midjourney terms, apparently translates to:
We removed it entirely. Hope that helps.
5/11
At some point I stopped trusting words.
That’s when I realized I had built the problem wrong.
I was trying to describe an anomaly into existence instead of stabilizing it visually.
So I did the only thing that actually works in this system:
I rebuilt the reference image.
Not as a reveal.
As a fact.
A girl in a factory uniform. Same world. Same lighting. Same exhaustion. But now with elf-like ears already present. Not magical. Not decorative. Just there. Like a genetic detail no one comments on.
That changed everything.
6/11
Lab note: Midjourney doesn’t handle “becoming” well.
It handles “already is.”
Once I understood that, the entire narrative shifted.
The story wasn’t:
“She discovers what she is.”
It became:
“He realizes what she is… and then what he is.”
That reversal matters.
Because the machine doesn’t like transformation arcs.
It prefers recognition arcs.
Things being discovered, not becoming.
7/11
I rebuilt the prompts.
Every single shot had to carry full identity description again.
No shortcuts.
No relying on memory between frames.
The model does not remember your intention. Only your repetition.
So the girl was always:
factory worker, fair skin, auburn hair, industrial jacket, work trousers, boots, fluorescent grime, overcast steel world.
The boy was always:
pale olive skin, dark wavy hair, hoodie, industrial jacket, work trousers, same world, same fatigue.
And every shot had to survive independently.
No narrative inheritance.
8/11
Then came the second correction.
The camera problem.
The machine wanted everything to be a portrait.
So I had to start breaking its instinct with blunt force composition language:
not facing camera
profile view
candid observation
unaware of being filmed
off-axis framing
It’s almost funny how primitive the solution is compared to the complexity of the intent.
You don’t negotiate with elegance.
You override with direction.
9/11
The turning point of the whole system was not emotional.
It was technical.
I stopped saying “subtle elf-like ears.”
Because Midjourney hears that as:
optional detail, ignore it.
Instead I had to treat it like furniture in the room.
Not symbolic.
Just present.
Like a chair.
Like a handrail.
Like something that exists whether the camera cares or not.
10/11
Lab note: the hug was the hardest frame.
Not emotionally.
Structurally.
Because Midjourney collapses physical closeness into generic romance geometry unless you actively resist it.
So even in the most intimate moment, I had to keep reminding the system:
work clothes
industrial dock
wet concrete
steam vents
no fashion styling
no romantic conversion
Otherwise it would turn grief and recognition into something like perfume advertising.
11/11
What I learned is not about elves.
It’s about constraint.
About how fragile narrative actually is when it passes through systems that do not understand narrative as continuity.
Only as isolated moments.
Each frame is its own truth.
So the story doesn’t live in the prompts.
It lives in the consistency of what you refuse to let the model forget.
And in the end, that’s what recognition is too.
Not transformation.
Not revelation.
Just two people realizing the same thing at the same time:
You were here the whole time.
Lab note: machines don’t understand recognition. But they can still show it.
12/11
Tools & Creative Stack
Midjourney — still image generation and character anchoring
Grok Imagine — image-to-video motion extension
Kdenlive — sequence assembly and timing
CchatGPT – AI-assisted prompt engineering
Image workflow automation (Midbot) — batch generation pipeline
13/11
Discovery / Takeaway
The machine doesn’t tell stories.
It stabilizes moments.
So if you want a story to exist inside it, you don’t build narrative.
You build repetition until narrative becomes the only thing that survives the repetition.
That’s the trick.
Not more imagination.
More consistency under pressure.
TL;DR: Midjourney is not a storyteller. It’s a pattern grinder. Feed it carefully.
14/11
Somewhere in all of this, the factory stopped being a factory.
It became a test of what holds when meaning is forced through systems that don’t believe in meaning.
And somehow, two small moments survived anyway.
A glance that didn’t become a portrait.
A silence that didn’t become empty.
Recognition, rendered in steel and fluorescent light.
Steve Teare
video alchemist
TerminallyBored.Monster
Palouse, Washington USA