If you’ve ever wondered how AI-generated content goes from concept to screen, buckle up—here’s a wild example of an idea moving from script to finished video in just a couple of days

Two weeks ago, Ali Powell, an AI video producer based in Tokyo, was tasked with creating a two-minute video for a company called iWallet. The brief was straightforward: produce a short, engaging video that highlighted iWallet’s latest product—a service that allows users to pay for home repairs quickly and effortlessly.

The video itself was fairly simple in structure, but the challenge was to make it lively, visually appealing, and informative all within a 90 second clip. The process started with the script, which laid out the key points about the product. It wasn’t Citizen Kane, but it could lend itself to something interesting.

Are you building an AI product? Work for a brand looking for AI video experts? We’re holding the first AI marketing video awards in New York this May! We’ll be posting more information this month but until then visit the website and sign up if you’d like to join as a brand sponsor.

Here’s a snippet of that script:

Powell started out the old way, shooting and cutting video for newsrooms like CoinDesk and TechCrunch. He knew how to build a story from raw footage, how to pace it, how to make it watchable and even viral. That part did not change. What changed was the ground under him. As AI tools crept into production, the market for that kind of hands-on work began to thin out. Some jobs vanished, others shrank, and the path forward stopped looking stable. He didn’t have to carry a camera and gear anymore and he felt lost for a while, wondering what his skills and background were good for.

So he shifted. The skills he already had like story sense, timing, and visual instinct, turned out to map well onto prompting. Instead of directing a crew, he began directing AI. Instead of waiting on budgets and schedules, he could move fast, testing ideas in hours that once would have taken weeks.

That is when things started to get strange, in a good way. Powell had always been deep into comics. With AI, he could finally build them. He began sketching out scenes that would never get greenlit in a studio system. Robin Williams as the Riddler, not as a joke but as a fully realized take on the character. William Katt, remembered by many as the lead in The Greatest American Hero, he recast as the Flash.

Powell was no longer limited by access or budget. He was limited only by taste and patience, which is how storytelling used to work before the machinery of production grew so large. This video hit 1 million watch hours just because it was so wild: a 1970s romp that mimics the visual style of old superhero shows almost perfectly.

With those skills in tow, Powell got to work.

“First, I generate stills with Nano Banana which set the scene and create the characters. For example, if I’m doing something with home repair I’ll generate a home owner and a contractor and maybe some kids. Just characters I’ll use over and over.”

Here’s the HVAC guy he made:

He’s careful to prompt properly.

“If you just give it junk, you get junk out,” he said. “You have to prompt very carefully.”

He sent this as evidence. Check out the dialogue:

The process continues with more and more detailed prompts. For example, he asked the AI to produce scenes with the HVAC repair person and the homeowner. The interactions are still a bit stilted but they get the point across. For this prompt he basically asked for his generated HVAC guy to meet the homeowner at their home.

And then wrote “The HVAC repair guy fixes the leak and replaces the unit, the female home owner happily pays him.”

Not all of the footage works on its own. There is always a trace of something off. Faces hold a fraction too long. Movements feel slightly delayed. A lot of it sits in that uneasy space where the image is close enough to feel real, but not close enough to forget how it was made.

That is where Powell’s background starts to matter. He writes dialogue to give the scenes shape. He cuts aggressively, trimming anything that drags or breaks the rhythm. What begins as something loose or even comedic starts to take on structure.

He works the way a traditional shooter would. He gathers more than he needs, then builds from it. He prompts for extra angles, small inserts, connective shots, the kind of B-roll that carries a story forward Then he weaves those pieces in between narration, letting the sequence breathe in a way that feels intentional rather than accidental. The result is not perfect, but it is coherent, and that still counts for a lot.

He could not share the final cut yet, but we will post it once it is ready. What he built, in practical terms, is a complete two minute commercial delivered in both portrait and landscape formats, finished in under two days. Not long ago, that kind of turnaround would have required a crew, a schedule, a location, and weeks of post production.

Now the work moves at a different pace. Powell can sketch, test, discard, and rebuild without waiting on anyone else. He can generate scenes, revise them, and slot them into place almost in real time. The process is faster, but it is also more fluid, closer to editing in the mind than coordinating a set.

He is not blind to what has been lost. There is a texture to traditional video, a physicality, a sense of presence that comes from real light, real movement, real people in a room. That does not fully carry over. He misses that. At the same time, he recognizes what he has gained. The range of ideas he can try has expanded. The cost of failure has dropped and he can follow an impulse all the way through to a finished piece, and that kind of creative freedom is hard to ignore.

“These tools give me superpowers,” he said.