There is a particular kind of creative problem I really enjoy.
I have a song that is already emotionally loaded, and my first instinct is to make a video that explains it.
Then I have to stop myself.
No.
Don’t explain it.
Let it breathe.
That became the guiding idea for my latest video project: Nothing Compares 2 U, rebuilt from several different vocal performances and wrapped in an entirely surreal visual world.
The song is melancholy. It is wistful. It has that strange feeling of wanting something that cannot simply be replaced.
So naturally, I decided not to show any of that.
No people.
No breakup scenes.
No lonely person staring through a rainy window.
No symbolic couple walking away from each other.
Absolutely none of the usual visual vocabulary.
Instead, I wanted impossible landscapes, strange objects, celestial reflections, flowers that shouldn’t exist, and places that feel as though I wandered into somebody else’s dream.
And then I had to figure out how to make those still images breathe.
2/9
Step 1: Start with the song.
The musical foundation came first.
I’m using vocal performances from four different artists:
Chris Cornell
Sinéad O’Connor
Lissie
Adam Levine
Rather than trying to make them sound like one singer, I decided to embrace their differences.
Different voices.
Different interpretations.
Different tempos.
Different keys.
That could have become a technical nightmare.
Instead, it started becoming part of the character of the piece.
The backing track I generated has a melancholy, ethereal quality. There is a little harmonic tension when the different vocals sit against it.
And you know what?
I liked it.
Lab note:
Sometimes the thing that looks like an error in the waveform is actually the thing giving the piece its personality.
I could have spent hours trying to eliminate every bit of dissonance.
Instead, I listened.
It sounded mysterious.
Keep it.
3/9
Step 2: Stop illustrating the lyrics.
This was probably the most important decision.
I deliberately did not want the visuals to correspond to individual lines.
When Chris sings about time passing, I don’t need a clock.
When somebody sings about loneliness, I don’t need an empty chair.
When the song says nothing compares to someone, I don’t need a picture of that person.
Those are illustrations.
I wanted something closer to emotional association.
The visuals should feel the way the music feels.
That’s different.
So I started building a visual vocabulary around five ideas:
- Impossible nature
- Cosmic landscapes
- Surreal objects
- Dreamlike details
- Pure atmosphere
The resulting shot list includes things like an ocean reflecting a moon that isn’t actually in the sky, a door standing in shallow water with no building attached to it, and a glass sphere containing an entire miniature night sky.
None of them “mean” anything specific.
That’s the point.
4/9
Step 3: Make the stills first.
This is where Midjourney comes in.
I’ve learned that I get much better results when I treat the still image as the actual artwork rather than treating it as disposable raw material for an animation.
The still has to work.
If I pause the video and look at one frame, it should be interesting enough to stand on its own.
So I built the shots as cinematic still photographs.
Wide landscapes.
Extreme macro details.
Bird’s-eye views.
Low angles.
Long lenses.
Lots of negative space.
And a consistent visual language:
deep blues, lavender, silver, violet, occasional warm light, atmospheric depth and surreal naturalism.
But there was another constraint.
I didn’t want the images to become funereal.
That’s an easy trap with this song.
Dark + lonely + slow music = somebody needs to call the undertaker.
Not what I wanted.
The emotional target became something more like:
wistful
mysterious
peaceful
beautiful
contemplative
quietly yearning
Melancholy, but alive.
5/9
Step 4: Thirty little dreams.
I ended up building a library of thirty surreal stills.
Some are enormous.
Some are tiny.
Some are almost cosmic.
Some are just one strange object sitting in the wrong place.
For example:
A gigantic pale moon hovering just above a dark ocean.
A single blue flower growing from a polished black stone.
A mountain reflected in a lake, except the reflected mountain is completely different.
A monumental stone arch with a star-filled sky visible through it.
A single pearlescent seashell with an impossible violet glow inside.
A sheet of ocean floating horizontally in an otherwise empty sky.
A black sun surrounded by a silver-lavender corona.
And one of my favorites: a completely empty ocean beneath an impossible lavender dawn, with one tiny warm point of light far beyond the horizon.
That’s the kind of image I want at the end.
Not an answer.
A feeling.
Lab note:
I am increasingly convinced that surreal imagery works best when you don’t explain the surrealism.
The moment you tell the viewer what the impossible thing represents, you’ve reduced it.
Let the viewer bring their own baggage.
I certainly have enough of mine.
6/9
Step 5: Now make the photographs breathe.
This is where Grok Imagine enters the picture.
And this is where I had to resist the temptation to over-prompt.
The still already knows what it looks like.
I don’t need to tell the animation model what color the sky is, what kind of tree is in the picture, where the moon is positioned, or what the camera is looking at.
The image has already done that job.
I just need to tell it how the image should behave.
My animation prompt is remarkably simple:
“subtle surreal cinematic dream, slow graceful environmental movement, slow cinematic camera push-in, subtle natural parallax, weightless movement, organic, hypnotic, emotionally restrained, ethereal fine-art cinema, ethereally beautiful, wistful, mysterious, peaceful, contemplative, quietly yearning”
That’s it.
No giant paragraph of instructions.
No twenty-seven contradictory commands.
No attempt to micromanage every pixel.
Lab note:
This is one of those places where AI has taught me something about directing.
Sometimes directing means telling the performer exactly what to do.
Sometimes directing means giving the performer a mood and getting out of the way.
For these images, I’m choosing the second approach.
7/9
Step 6: Let the world move, not the story.
The animation is intentionally subtle.
I don’t want the camera racing through a surreal landscape like it just discovered a video game.
I want the image to breathe.
A little atmospheric movement.
A little parallax.
A slow push toward the subject.
Perhaps water moving almost imperceptibly.
Perhaps fog shifting.
Perhaps a reflection changing.
Perhaps a tiny change in light.
The viewer shouldn’t necessarily notice the animation immediately.
They should simply feel that the photograph is alive.
That distinction matters.
The still image is the dream.
The animation is the breathing.
The music is the emotion.
And I’m sitting in the middle of all three, wearing my video-alchemist hat and trying not to break anything.
8/9
The tools on the workbench
The current creative stack for this project looks something like this:
Music generation and experimentation — Google Flow Music AI
Vocal separation — Lalal.ai
Still-image generation — Midjourney
Image-to-video animation — Grok Imagine
Audio editing and timing — Audacity
Video assembly and finishing — Kdenlive
And, of course, ChatGPT — because apparently I have reached the stage of my life where I argue with a language model about whether a visual prompt contains too many words.
It does.
It usually does.
The trick is knowing when to stop adding them.
The Real Lesson
The most interesting thing about this project isn’t really the technology.
It’s the decision to leave room for ambiguity.
I started with a song that carries a very specific emotional history.
It would have been easy to make a video that told the audience exactly what to feel.
Instead, I’m trying to create a visual environment where they can feel whatever they bring into it.
That’s a much more interesting collaboration.
The AI gives me things I wouldn’t necessarily have imagined on my own.
Sometimes they’re exactly right.
Sometimes they’re ridiculous.
Sometimes they are technically wrong and emotionally perfect.
Those are my favorite accidents.
The master alchemist’s job isn’t to prevent accidents.
It’s to recognize which accidents are worth keeping.
TL;DR:
Don’t illustrate the emotion.
Build a world that has the same weather as the emotion.
Then let it breathe.
9/9
I think that’s what I’m really chasing with these projects now.
Not realism.
Not perfection.
Not even necessarily narrative.
I’m interested in atmosphere.
A feeling that exists for a few minutes and then disappears.
A landscape that could not exist, but somehow feels familiar.
A piece of music that carries several different voices without forcing them to become one.
And a collection of pixels that, with a little help from some algorithms, begins to feel almost alive.
That’s the magic trick.
Not making AI create something by itself.
Not making myself create everything by hand.
The interesting place is somewhere in between.
I provide the intention.
The machines provide the surprises.
Then I decide which surprises belong in the dream.
That’s the work.
That’s the fun.
And, occasionally, that’s where the art sneaks in.
Steve Teare
video alchemist
TerminallyBored.Monster
Palouse, Washington USA