Case study · AI music video · 2026
Case study · AI music video · 2026
To The Moon
To The Moon
To The Moon
One director and an AI producer made a 75 second music video in two days. Here's what that means for creative teams.
One director and an AI producer made a 75 second music video in two days. Here’s what that means for creative teams.
2
2
days, brief to graded master
days, brief to graded master
$137
$137
in image and video generation (USD)
in image and video generation (USD)
38
38
shots in the cut, from 280 stills and 107 clips
shots in the cut, from 280 stills and 107 clips
10
10
full cuts before picture lock
full cuts before picture lock
Overview
Overview
One director and an AI producer made a 75 second music video, with an original song, in two days for about $137 in generation costs. I directed every creative decision. Claude ran production: writing, prompting, timing, code, review tooling and edit assembly. Specialist models made the images, video and song.
One director and an AI producer made a 75 second music video, with an original song, in two days for about $137 in generation costs. I directed every creative decision. Claude ran production: writing, prompting, timing, code, review tooling and edit assembly. Specialist models made the images, video and song.
I didn't do this to prove AI can make a video. I did it to find out how a creative lead should work when production is no longer the bottleneck. This is what I learned.
I didn’t do this to prove AI can make a video. I did it to find out how a creative lead should work when production is no longer the bottleneck. This is what I learned.
01 · Baseline
01 · Baseline
Same director, one year apart
Same director, one year apart
Same director, one year apart
A year earlier I made "Jenna Ortega" for the artist Stellar. It won the Artist's Choice Award at the 2025 OpenArt Music Video Awards. It was the same kind of brief: an original song, a lead character, a world, and a story with a turn. What changed was the process.
A year earlier I made “Jenna Ortega” for the artist Stellar. It won the Artist’s Choice Award at the 2025 OpenArt Music Video Awards. It was the same kind of brief: an original song, a lead character, a world, and a story with a turn. What changed was the process.
2025 · Jenna Ortega
2025 · Jenna Ortega
2026 · To The Moon
2026 · To The Moon
Length
Length
2:04
2:04
1:15
1:15
Time
Time
About 10 days
About 10 days
About 2 days
About 2 days
Generation cost
Generation cost
About $200 in credits (sponsored)
About $200 in credits (sponsored)
About $137
About $137
Images
Images
2,000+, generated by hand
2,000+, generated by hand
About 280, generated by Claude, approved by me
About 280, generated by Claude, approved by me
Video clips
Video clips
200+, generated by hand
200+, generated by hand
About 107, generated by Claude, 38 used
About 107, generated by Claude, 38 used
Working with AI
Working with AI
One prompt at a time, every generation by hand
One prompt at a time, every generation by hand
I gave direction, Claude ran production end to end
I gave direction, Claude ran production end to end
My role
My role
Every hat: writer, art department, generator, editor
Every hat: writer, art department, generator, editor
Director
Director
It isn't a like for like test. The 2025 film is longer, and it was made under competition rules with sponsored credits. Per second of finished film, it's still about three times faster.
It isn’t a like for like test. The 2025 film is longer, and it was made under competition rules with sponsored credits. Per second of finished film, it’s still about three times faster.
Speed isn't the biggest change, though. In 2025 I carried every box myself. In 2026 the AI is the crew, and my hours go into decisions. The taste that won the first award is what directed the second film.
Speed isn’t the biggest change, though. In 2025 I carried every box myself. In 2026 the AI is the crew, and my hours go into decisions. The taste that won the first award is what directed the second film.
02 · Why it matters
02 · Why it matters
Who this is for
Who this is for
Who this is for
CMOs
CMOs
The cost of an idea has collapsed, and the cost of a bad idea has too. Ten full versions in two days means you can test creative the way you test copy. The question is no longer "can we afford to produce this?" but "do we know what we want to say?"
The cost of an idea has collapsed, and the cost of a bad idea has too. Ten full versions in two days means you can test creative the way you test copy. The question is no longer “can we afford to produce this?” but “do we know what we want to say?”
Creative directors
Creative directors
Taste is now the bottleneck, and that's good news for us. Every hour I'd normally spend keyframing went into direction: the idea, the casting, the performance notes, which take, which frame, when to stop.
Taste is now the bottleneck, and that’s good news for us. Every hour I’d normally spend keyframing went into direction: the idea, the casting, the performance notes, which take, which frame, when to stop.
AI creators
AI creators
The edge isn't the prompt. Prompts and models change every month. What lasts is the system around them: a written bible, a measured edit, approval gates, and code for anything that has to be exact.
The edge isn’t the prompt. Prompts and models change every month. What lasts is the system around them: a written bible, a measured edit, approval gates, and code for anything that has to be exact.
03 · Operating model
03 · Operating model
A studio with two seats
A studio with two seats
A studio with two seats
Director · Me
Director · Me
The idea, the look, casting, every approval, the edit decisions, the grade, the final 10%
The idea, the look, casting, every approval, the edit decisions, the grade, the final 10%
Producer · Claude
Producer · Claude
Brief and canon docs, lyrics and song prompts, image and video prompts, the beat map, Python graphics, the review tool, Premiere assembly, the After Effects build script
Brief and canon docs, lyrics and song prompts, image and video prompts, the beat map, Python graphics, the review tool, Premiere assembly, the After Effects build script
Specialists
Specialists
Suno, Seedream, Nano Banana, Seedance, Kling (via the Leonardo API), ChatGPT. Song, stills, surgical fixes, motion and lip sync, a first pass colour grade.
Suno, Seedream, Nano Banana, Seedance, Kling (via the Leonardo API), ChatGPT. Song, stills, surgical fixes, motion and lip sync, a first pass colour grade.
The model doesn't depend on a particular tool. The producer seat orchestrates whichever specialist models are best this month. Swap any of them and the structure still holds.
The model doesn’t depend on a particular tool. The producer seat orchestrates whichever specialist models are best this month. Swap any of them and the structure still holds.
04 · The production system
04 · The production system
Six principles that transfer to brand work
Six principles that transfer to brand work
Six principles that transfer to brand work
1
1
Brief for meaning, not output
Brief for meaning, not output
The brief centred on one phrase doing three jobs. "To the moon" is euphoria, a valuation going up and a literal launch. Every shot hangs off it. AI will happily generate a hundred directions. A sharp brief is what makes the one you pick mean something.
The brief centred on one phrase doing three jobs. “To the moon” is euphoria, a valuation going up and a literal launch. Every shot hangs off it. AI will happily generate a hundred directions. A sharp brief is what makes the one you pick mean something.
For brands: the single-minded proposition matters more now, not less.
For brands: the single-minded proposition matters more now, not less.
2
2
Turn the brand into a system the machine reads
Turn the brand into a system the machine reads
Before any image, Claude wrote a brief, a canon document (character, wardrobe, props, colour rules, what never appears) and a storyboard mapped to the music. Then we locked the world as reference images: a face anchor, the hero suit, a character turnaround, the sets. Every prompt read those first.
Before any image, Claude wrote a brief, a canon document (character, wardrobe, props, colour rules, what never appears) and a storyboard mapped to the music. Then we locked the world as reference images: a face anchor, the hero suit, a character turnaround, the sets. Every prompt read those first.
For brands: your guidelines, rewritten so a model can follow them.
For brands: your guidelines, rewritten so a model can follow them.
3
3
Measure the spine
Measure the spine
The song was measured into data: 134 BPM, 42 bars, every beat and every sung word timestamped. Cuts, captions, flashes and camera shake were all placed from that data, not by ear.
The song was measured into data: 134 BPM, 42 bars, every beat and every sung word timestamped. Cuts, captions, flashes and camera shake were all placed from that data, not by ear.
For brands: measure the VO or music bed once. Every cutdown inherits the timing.
For brands: measure the VO or music bed once. Every cutdown inherits the timing.
4
4
Approve cheap, then spend
Approve cheap, then spend
Every shot was approved as a still (cents) before it was animated (around $1.30 to $1.60 a clip). When something broke, like an extra finger or hair outside the helmet, we fixed that frame surgically instead of regenerating, and kept what was already right.
Every shot was approved as a still (cents) before it was animated (around $1.30 to $1.60 a clip). When something broke, like an extra finger or hair outside the helmet, we fixed that frame surgically instead of regenerating, and kept what was already right.
For brands: put the approval gate where changes are cheap.
For brands: put the approval gate where changes are cheap.
5
5
Anything that must be exact is code
Anything that must be exact is code
The HUD, radar, countdown, warning bars and titles are not AI generated. Claude wrote them in Python: an original dot matrix font defined as data, plus small generators for each graphic. A change like "make the countdown say IMPACT" was one line and a 4K re-render.
The HUD, radar, countdown, warning bars and titles are not AI generated. Claude wrote them in Python: an original dot matrix font defined as data, plus small generators for each graphic. A change like “make the countdown say IMPACT” was one line and a 4K re-render.
For brands: type, logos, legal lines and end frames should never be left to a diffusion model.
For brands: type, logos, legal lines and end frames should never be left to a diffusion model.
6
6
Govern the review
Govern the review
After the first cut, Claude built a review page for the project. The cut plays alongside a card for every shot. I marked each one Keep, Re-roll or Trim with a reason, and Claude read the decisions back and acted on them. Ten rounds, no lost notes.
After the first cut, Claude built a review page for the project. The cut plays alongside a card for every shot. I marked each one Keep, Re-roll or Trim with a reason, and Claude read the decisions back and acted on them. Ten rounds, no lost notes.
For brands: a decision log tied to every asset lets a small team ship fast without losing control.
For brands: a decision log tied to every asset lets a small team ship fast without losing control.
The finish stays human
The finish stays human
AI got to picture lock fast. The last 10% was craft: frame trims in Premiere, an After Effects build that Claude scripted to the beat map, and the grade. ChatGPT did a first pass grade on 41 clips and a signal glitch effect, with everything left editable. Then I took over, pushed the glitch further and finished it by hand.
AI got to picture lock fast. The last 10% was craft: frame trims in Premiere, an After Effects build that Claude scripted to the beat map, and the grade. ChatGPT did a first pass grade on 41 clips and a signal glitch effect, with everything left editable. Then I took over, pushed the glitch further and finished it by hand.
05 · Guardrails and limits
05 · Guardrails and limits
What I held to, and what still needs people
What I held to, and what still needs people
What I held to, and what still needs people
Guardrails
Guardrails
No real person's likeness and no third party logos, in prompts or outputs. A human approved every still, every clip and every cut. AI use is disclosed on publication. No credentials or posting handed to the AI. I publish myself.
No real person’s likeness and no third party logos, in prompts or outputs. A human approved every still, every clip and every cut. AI use is disclosed on publication. No credentials or posting handed to the AI. I publish myself.
Where it still needs people
Where it still needs people
Faces drift without a locked reference on every generation. Hands and crowds still break. Scale in wide shots goes wrong easily. And taste: the model offers, it doesn't know when something is right. That call was mine every time.
Faces drift without a locked reference on every generation. Hands and crowds still break. Scale in wide shots goes wrong easily. And taste: the model offers, it doesn’t know when something is right. That call was mine every time.
What this means for creative teams
What this means for creative teams
Craft moves up the chain.
Craft moves up the chain.
Craft moves up the chain.
A small group with clear direction, a written brand system and AI producers can now make work that used to take a studio. That isn't the end of craft. It moves craft up the chain, to the brief, the system, the approvals and the finish.
A small group with clear direction, a written brand system and AI producers can now make work that used to take a studio. That isn’t the end of craft. It moves craft up the chain, to the brief, the system, the approvals and the finish.
The process changed almost completely in twelve months. The idea, the taste and the storytelling didn't, and they're still what wins.
The process changed almost completely in twelve months. The idea, the taste and the storytelling didn’t, and they’re still what wins.
Bring this to your team →
© 2026 Phyo Thu · @phyofox
Wellington, NZ
© 2026 Phyo Thu · @phyofox
Wellington, NZ