AI Video: Magic Button or Just Another Tool in the Toolbox?
We Take You Behind the Scenes of a Project Built Entirely With AI
There’s a common belief right now that Artificial Intelligence is making creative work faster, easier and more efficient. And in many ways, it is. Most people have tried it. You type a prompt, hit enter, and something impressive appears.
Job done. Easy!
But in the creative industries there’s also a quiet fear humming in the background…
'Will AI replace jobs?'
'Will it make traditional craft obsolete?'
'Are we heading toward a world where no one needs to go out and physically make anything anymore?'
We’ve just completed a project that gave us the chance to test those ideas for real. And we feel like we came away with some lessons worth sharing.
We’ve partnered with a marketing and design agency that we’ve worked with for years. They brought us a brilliant challenge: their client wanted a four-part fictional series set in the not-too-distant future. Each episode followed a different global brand and imagined how AI might shape its future.
The brands weren’t real, but they were close enough to reality to feel believable.
The brief was to create four three-minute films in a documentary style, generated entirely with AI visuals and AI voice. We even suggested AI-generated interviews to hear directly from “CEOs” and customers.
Could we have filmed it for real? In theory, yes.
In practice it would have meant shooting across multiple countries, building fictional stores, designing future worlds and layering heavy VFX on top. You’re comfortably into six figures, probably more. Without a feature-film budget, it simply wasn’t realistic.
Stock footage didn’t get close either. The narrative needed moments that didn’t exist.
AI was the only viable route.
What we didn’t expect was just how much human craft would still be required.
The First Big Hurdle: Consistency
In a normal shoot, you cast an actor and they stay the same from shot to shot. With AI, every generation is a roll of the dice. Faces change. Clothes change. Backgrounds mutate.
So before we generated any video, we had to build a system.
We began with the supplied scripts and created a written storyboard. For every line we mapped out a visual prompt: what’s happening, where we are, who’s there. Then we layered in filmmaking language; camera type, lens, lighting style, mood, production design.
Once that was locked in, we generated still images to create a visual storyboard and, crucially, to “cast” our characters.
Getting those people to remain recognisable across an entire film took serious fine-tuning. Tiny prompt adjustments, regenerations, back and forth, over and over.
When we finally had a stable set of characters and environments, those stills became the reference for image-to-video generation.
Only then did we start to move.
Prompt. Generate. Fix. Repeat.
This is where the myth of the magic button fades.
You prompt.
You generate.
You notice the extra arm, or a person morphing into furniture, or something odd happening in the background.
You tweak and go again.
And again.
Sometimes we regenerated a single moment more than 50 times.
Keep the same prompt and you’ll get variations. Change a word and you might fix one problem while creating another. A lot of the time it felt like potluck.
What’s fascinating is how convincing it looks at first glance. On first watch you think, wow, that’s incredible. Watch it three or four times and the cracks start to show. Movements that feel off, objects behaving strangely, tiny things that break reality.
Even many of the shots we approved still contained minor imperfections. At some point you accept that absolute perfection just isn’t available yet.
Text was another battle. AI really struggles with legible language. If a shop sign, product label, animation pop-out etc needed to read properly, we exported the still into Photoshop, designed the text manually, and then fed that corrected image back into the generator.
New technology, traditional skills. Side by side.
The Interview Problem
The scripts relied heavily on interviews; talking heads that would give the films authority and a documentary feel.
Simple, right?
Not quite.
We tested platforms offering automatic lip-sync and dubbing. Technically amazing compared to even a year ago, but still not convincing enough when you’re aiming for broadcast-quality storytelling. Mouths warped. Expressions drifted.
So again, we went manual.
First we generated the characters as stills.
Then we asked the video tools to create them speaking the lines within the prompt.
But instead of using the built-in voices, we produced our preferred dialogue separately with AI voice tools. We needed specific accents and ethnicities to match the visuals and this was the only way to ensure consistency.
Finally, we retimed and dubbed everything manually to match in the edit.
Why the extra pain? Because we needed consistency. If the CEO appears at the start and again at the end, the voice must match perfectly. Generating audio inside each clip produced subtle differences that an audience would absolutely notice.
It’s amazing tech.
It just isn’t controllable enough yet.
The Edit: Saving the Good Bits
Even after approvals, we couldn’t drop clips onto the timeline and export.
We masked backgrounds.
We painted out artefacts.
We hid strange details.
“Why not regenerate?”
Because you’ll never get the same performance twice.
If a delivery, expression or camera move worked, but there was a glitch in the corner, it was far safer to repair the issue than gamble on losing the magic entirely.
One interview gave our character a row of mystery buttons down a polo shirt. Everything else was perfect. So we removed the buttons.
Problem solved.
So, The Verdict?
AI allowed us to produce films that otherwise wouldn’t have existed without enormous budgets. That’s extraordinary.
But fast and easy? Not even close.
The project leaned heavily on fundamentals: storytelling, shot design, editing, colour grading, pacing, problem-solving. The tools opened the door, but they didn’t walk us through it.
AI is powerful.
But it is still a tool.
And like any tool, results depend on the people using it.
What’s also clear is how quickly the landscape shifts. During the three months we worked on this series, some capabilities improved dramatically, while others were restricted by platform owners because of deep-fake or copyright concerns. The ground keeps moving.
Maybe one day the magic button arrives.
For now, if you want great work, you still need a great team.
If you’re exploring how AI might fit into your video strategy, just know this: the button isn’t magic yet, but in the right hands, the outcomes can still be pretty special.
