I’ve produced hundreds of hours of television. Nothing prepared me for this one.
Fifteen years in TV. I know how stories work. I know pacing, structure, the difference between a scene that lands and one that doesn’t. I’ve worked with crews, with talent, with executives who had opinions about everything.
I wasn’t new to AI video tools either. I’d made a commercial, played around a lot. But I’d never had this much riding on the output. This wasn’t just a gig. It was for the team my husband (Erik Deutsch) co-founded. Their first impression on the world.
SmartFocus.ai needed a brand film. Not a product demo. Not a feature walkthrough. A film that said this is who we are before anyone clicked a single button. And the centerpiece of that identity was FocusFox, the AI guide that already lived on their website. The character existed. My job was to figure out who she was in a film. How she talked. What she wanted to say. And how she wanted people to feel when they clicked away.
Turns out that’s not a small job.
I’m only half joking. When you’re building a character for a brand film you have to make decisions that sound ridiculous out loud. Does she work at the museum? Is she a visitor? Does she walk on all fours or does she get a lanyard and a blazer?
First question: does she have a jacket?
I’m only half joking. When you’re building a character for a brand film you have to make decisions that sound ridiculous out loud. Does she work at the museum? Is she a visitor? Does she walk on all fours or does she get a lanyard and a blazer?
I went through a lot of iterations. Each script went back to the team, who are sharp professional communicators, and came back with notes. They pushed back. I pushed back. That friction was the most valuable part of the process, even when it was maddening, because it forced every decision to earn its place.
What we landed on felt right the moment we got there. She just knows this place. She’s been here before. She’s not surprised by any of it. Moving on.
The film is called The Museum of Confident Mistakes.
New Coke. Harley Davidson cologne. The Tropicana rebrand. A fox who walks through history’s most confident brand decisions and delivers the tour with the energy of someone who saw it coming.
The joke, if you can call it that, is that none of these companies thought they were making a mistake. They were certain. They just never asked the right questions. That’s what SmartFocus is for. And that’s what FocusFox needed to embody: warm, knowing, the kind of guide who helps you ask the questions you didn’t know you needed to ask. The museum was the perfect world for her to live in. Because nobody wants to end up in it.
Here’s what I learned building it.
Lesson one: lock the voice before you touch anything else.
My instinct was to build the visuals first. That’s how production works. Instead I ran the full script through ElevenLabs and locked the voiceover before I built a single image. Smartest decision I made.
Here’s why: if you change even one line after the fact, you redo the entire VO. You cannot patch AI voice recordings. The moment you stitch together separate takes, something shifts. It stops sounding like a character and starts sounding like software. Locking the voice first forced me to be certain about the script before I committed to anything else. That discipline made everything downstream better.
Lesson two: start with stills, not video.
I built the Fox and each museum exhibit as images first, in Google Flow and ChatGPT, then imported everything into Higgsfield Soul 2.0 to lock consistency across every location. Then I dropped those stills into iMovie alongside the locked audio before generating a single clip. Credit where it’s due: my friend Tim Street suggested this. He runs an AI Filmmakers group that I tapped into every week throughout production, and it was one of the best resources I had. Exactly right, Tim. Suddenly I could see where everything lived, how long each beat needed, where the edit worked and where it didn’t. It also told me exactly how long each clip needed to be, which meant I wasn’t burning credits guessing. Low tech. High value. Do this first.
Lesson three: get feedback before you build, not after.
Before I generated one clip I ran the script through a SmartFocus focus group. Partly because I wanted the feedback. But also because I wanted to push the platform and see how it would handle the subject matter. It turned out to be one of the more interesting parts of the whole process.
Screenwriters first, to pressure-test the writing. Then directors and cinematographers to think through the visual language and how to keep the museum consistent across exhibits.
I built more exhibits than we ended up using. One was a community center that got torn down and never replaced, and how the town never recovered. The Spike Lee AI persona told me to keep it no matter what. He made a compelling case. But the team was right to push back. You don’t want people walking away from a brand film feeling demoralized. You want them feeling like they just found the thing they needed. The exhibit was good. It just wasn’t right for this.
What I loved was that the personas argued their corners like real collaborators. They didn’t all agree with each other. That’s not a bug. That’s exactly what good feedback looks like.
When the film was picture-locked I brought in an AI focus group of music supervisors to figure out what the score needed to be.
I was using the focus group tool to make the focus group tool’s film. I know how that sounds. But getting feedback before I was deep into production meant I wasn’t tearing apart things I’d already spent hours building. That lesson applies whether you’re using AI tools or not.
Lesson four: learn the platform before you spend the money.
Higgsfield was running a free Seedance launch when I started generating video. Eight second clips at no cost while I figured out the platform. I used that window hard. By the time I moved to Kling 3.0 for the actual production I knew what I was doing. More control, more economical, and for a film living online 1080p was exactly right.
Eight seconds at a time. One action per clip. Try to do too much in a single prompt and the output gets muddy. Keep it simple, move forward only when what you have is right.
What I didn’t expect:
How much I’d care about getting her right. Not the tools. Not the workflow. Her. Whether she reads as certain versus cold. Whether the moment lands or just sits there. That’s a character problem. And it turns out I’ve been solving character problems my whole career. I just never had to do it eight seconds at a time.
Brand storytelling isn’t about showing what your product does. It’s about making people feel what it’s like to be in the room with it. Every choice about FocusFox is a product decision as much as a creative one. You’re building someone people will remember long after they’ve forgotten the feature list.
The Museum of Confident Mistakes is live now on SmartFocus.ai. Go watch a fox walk through the worst decisions in brand history and deliver the tour with complete equanimity.
She already knew how it was going to end. That’s kind of the point.
Babette makes things. Television. Communities. AI tools. Occasionally, museums. You can learn more about her at pepajmedia.com. She is also the very proud wife of one of the brilliant people behind SmartFocus.ai. Go see what they built. You won’t end up in the museum for it.