“Well I’ll be damned.
Here comes your ghost again.”
Some songs don’t tell stories.
They summon them.
A few days ago I found myself listening to Joan Baez’s Diamonds and Rust for the hundredth time. Maybe the thousandth. At seventy years old, songs don’t arrive alone anymore. They drag entire decades behind them.
As the lyrics unfolded, I stopped hearing a song.
I started seeing a film.
Not a music video.
A memory.
A ghost story disguised as nostalgia.
And like most of my projects these days, I immediately decided to make my life harder by turning the whole thing into an AI-generated vertical short.
Because apparently I enjoy suffering.
2/9
The spark: one phone call and ten lost years.
The image that grabbed me wasn’t the famous chorus.
It was the payphone.
A man standing somewhere in the Midwest in 1974.
A pocket full of coins.
A phone number he hasn’t dialed in ten years.
That image felt cinematic.
The more I thought about it, the more I realized the entire song hangs on that single moment.
Not the romance.
Not the breakup.
The hesitation.
The decision to dial.
The possibility that hearing one voice could collapse an entire decade into a few seconds.
That’s where the project started.
Everything else grew from there.
Lab note:
The best AI film projects usually begin with one image, not one plot. Find the image that won’t leave you alone and follow it.
3/9
Building the hidden story.
The song is written from Joan Baez’s perspective.
But I became fascinated by the man on the other end of the line.
What if he told his side?
What if the call wasn’t really about reconnecting?
What if it was about mourning?
Not mourning her.
Mourning who he used to be.
That idea became the one-minute voiceover that follows the song.
The key line ended up being:
“I wasn’t nostalgic. Not exactly. I was mourning a version of myself that only existed when you were there to see me.”
That sentence unlocked the entire project.
Suddenly this wasn’t a love story.
It was a story about identity.
4/9
Step 1: Designing the visual language.
I wanted the entire piece to feel like 1974.
Not Hollywood’s version of 1974.
Actual 1974.
Brown corduroy jackets.
Suede coats.
Payphones.
Train stations.
Coffeehouses.
Winter breath hanging in the air.
The visual strategy became surprisingly simple:
• Present day = loneliness
• Memory = warmth
• The phone call = bridge between worlds
The memories needed to feel slightly dreamlike.
Not fantasy.
Just the way memory softens edges and edits reality.
The present-day shots needed more distance.
More silence.
More empty space.
A lot of storytelling happens between images rather than inside them.
5/9
Step 2: Creating the shot architecture
The final sequence ended up containing twenty-nine primary shots and fifteen additional insert shots.
Forty-four images total.
That may sound excessive.
It isn’t.
Three-minute montage films eat images faster than popcorn.
Some of my favorite visual anchors included:
• The moon above a rural payphone
• Robin-blue eyes
• Tarnished cufflinks
• Snowfall in Washington Square
• Motel signs beside forgotten highways
• Breath clouds merging in winter air
• An abandoned payphone receiver hanging in silence
Lab note:
If an object appears more than once, it becomes a symbol.
If it appears three times, it becomes a theme.
The payphone became the film’s emotional center of gravity.
6/9
Step 3: Composing the soundtrack.
The music prompt ended up being almost comically specific:
“Instrumental, felt piano, sparse minimalist arrangement, intimate and reflective, warm analog ambience, soft evolving synth pads, cinematic melancholy, moonlit solitude, nostalgic longing…”
And so on.
The interesting thing wasn’t the music itself.
It was how the music changed the visuals.
As soon as I heard the first drafts, the film slowed down.
Shots got longer.
Camera movement became gentler.
Images lingered.
The soundtrack wasn’t supporting the visuals.
The visuals were obeying the soundtrack.
That’s something AI projects constantly teach me.
The tools collaborate whether you planned it or not.
7/9
Tools & Creative Stack
Visual Design
• Midjourney
• ChatGPT
Narrative Development
• ChatGPT
• Human memory
• Regret
• Time
Audio
• AI music generation
• Voiceover narration
Editing
• Kdenlive
Project Specs
• Vertical format
• Approximately 3 minutes duration
• 44 generated images
• 1974 visual aesthetic
• Romantic drama style
• No dialogue except narration
• Mood-driven pacing
Budget
Mostly time.
Which, at seventy, is arguably the most expensive thing in the room.
8/9
The real lesson
The biggest surprise wasn’t technological.
It was emotional.
When I started this project, I assumed it was about lost love.
By the end, I wasn’t so sure.
The older I get, the more I suspect nostalgia isn’t really about another person.
It’s about versions of ourselves that disappeared along the way.
Certain people become mirrors.
When they leave, the reflection leaves too.
Years later we think we miss them.
Sometimes we do.
But sometimes we’re searching for the person we were when they looked at us.
That’s what this film became.
Not a romance.
A conversation with a ghost.
And the ghost wasn’t her.
9/9
TL;DR
I started with a Joan Baez song.
I ended up with a three-minute AI-generated meditation on memory, identity, and a payphone sitting beneath a full moon.
Along the way I generated forty-four images, wrote a fictional monologue, built a soundtrack, and accidentally discovered that nostalgia and grief are distant cousins.
Funny how that happens.
The machines helped me build the film.
But they didn’t provide the ghost.
That part was already waiting for me.
Steve Teare
video alchemist
TerminallyBored.Monster
Palouse, Washington USA