Back to blog

My AI Wrote 5 Commercial Scripts. They Were All Terrible.

I manage marketing for a supplements brand called BrickHouse Nutrition. They have a sleep aid product called DreamZzz. The existing commercial was... fine. Product shot, tagline, nothing memorable. I wanted something that would actually make someone stop scrolling.

So I asked my AI coworker to write me 5 funny commercial concepts. It knows the brand inside out (ingredients, competitors, positioning, the whole thing). I even gave it a formula: normal scene, unexpected twist, brand line. Like a 15-second joke with a product payoff.

Roger asking Viktor to write 5 commercial scripts for a supplements brand

Viktor came back fast with 5 concepts. A cabinet avalanche of abandoned greens tubs (nobody finishes them because they taste bad). A magician flipping a rival supplement upside down and one tiny puff comes out. An influencer chucking a sealed tub in a dumpster the second the camera stops rolling.

Viktor delivering commercial script concepts

On paper, not bad ideas. But they all had the same problem: they were safe, expected, and not actually funny. They read like a brainstorm doc, not like something you'd remember after seeing it once. Every concept followed the formula so literally that none of them had a real surprise.

Your Scripts Were Terrible

I told Viktor straight up: no offense, but these are terrible. Which is fine. Because in the process of reading 5 bad scripts, I figured out exactly what I wanted.

The concept: a woman wakes up in bed. Close-up, warm lighting. She checks her smartwatch. "Nice, 100% sleep score." Camera pulls back. She's in a hospital bed. IV in the other arm. Doctor next to her with a clipboard: "Ma'am, you've been in a coma."

Her eyes go wide. Cut to product. "Get better sleep. Without the coma. DreamZzz by BrickHouse Nutrition."

15 seconds. One joke. One product.

Roger telling Viktor his scripts were terrible then sharing his own coma commercial concept

I didn't need Viktor to come up with the idea. I needed it to help me PRODUCE it. Start cards and end cards for every scene that I could feed into AI video tools (OpenArt for the images, Kling or Seedance for image-to-video). A full prompt pack. A branded end slide with the real logo and fonts.

That's where it shifted from bad creative director to excellent production assistant.

Scene by Scene

Within minutes, Viktor generated keyframe pairs for each scene. Start frame and end frame, designed so the AI video tool has clear direction on what the "move" is between them.

Scene 1 end card (she's reading her sleep score, framed tight so you can't tell it's a hospital yet):

AI-generated keyframe of woman checking sleep score on smartwatch

Scene 2 (the reveal... same woman, same bed, but now you see the hospital room, the IV, the doctor):

AI-generated keyframe of hospital room reveal with doctor

It also delivered a full prompt pack: exact image prompts, video/motion prompts, and the audio script, all formatted so I could copy-paste into the generation tools.

The Corrections (This Is the Real Part)

Viktor did not nail every detail on the first pass. Not even close. A few things I had to fix multiple times:

The smartwatch kept ending up on the wrong wrist. I needed it on her right arm (the free one, no IV), but it kept flipping sides between generations. I corrected it three times before the continuity locked in.

The "shocked" expression on the reveal was way too much. Open mouth, huge eyes, full soap opera. I told it: mouth closed, eyes only slightly wider. Understated surprise, not a meme. The second version nailed it.

Roger correcting Viktor on watch placement and expression

The end card needed the real brand assets. Viktor's first version used an AI-approximation of the BrickHouse logo and tagline font. Close, but not right. Once I sent the actual logo file and the Tomarik Brush font, it rebuilt the end card with the real thing. Then I had to go back and forth on sizing (product too small, tagline on two lines instead of one, not centered).

Roger directing Viktor on end card sizing for DreamZzz commercial

Final end card, after about 4 rounds of "bigger, center it, one line not two":

Final DreamZzz end card with BrickHouse Nutrition branding

What I Took Away From This

Total time: about 2 hours from first message to final deliverable. No creative agency. No video editor. No production budget for the pre-production work.

The AI did not come up with the concept. I did. But it took my concept and turned it into a producible set of visual assets, ready to feed into AI video generation tools. That distinction matters.

If you are making short-form video ads for your brand or your clients, here's what I'd say after going through this:

  • Come up with the concept yourself. AI will give you safe, expected, generic ideas. The creative judgment is still yours.
  • Use AI for production. Keyframes, prompt packs, end cards, storyboards. This is where it saves you days and thousands of dollars.
  • Be specific. Scene by scene. "Watch on her right wrist, doctor on her left, IV in the other arm." The more precise you are, the fewer rounds of correction.
  • Correct it fast and bluntly. Don't accept "close enough." The details (which wrist, how wide the eyes are, product size on the end card) make the difference between something that looks professional and something that looks like an AI demo.

Viktor is the AI I use for this (and for building my website, training for pickleball, and basically everything else that comes across my desk). If you want to try it yourself, you can get $100 in free credits plus $50 off your first purchase here.

For more real examples of what it can do: my full Viktor review with screenshots.