My first AI video took 4 hours and cost $25

prompts + exact process, my Veo 3 AI grandparents for B2B marketing, lessons learned.

Jul 24, 2025

On Tuesday I posted my first AI video [watch here].

In this newsletter I'm sharing my exact process, all the prompts, and what I learned spending an afternoon with Google's Veo 3.

My video features 7 different AI grandparents reading Gen Z slang while roasting spray-and-pray marketing. Of course.

I have ZERO video skills. But after seeing what people are creating with AI video, I had to try it myself.

My goal: Create a 60-second "commercial" for Keyplay that *might* make people stop scrolling and would definitely help me learn about AI video.

Here's how I got there and what I learned.

The Setup (15 minutes)

First, I created a Claude project and loaded it with two types of knowledge:

Veo 3 examples and prompts I found online (links at the end of this post)
Some of my past LinkedIn posts about B2B marketing and Keyplay positioning

This saved me from repeating context with each iteration. Highly recommend this simple step.

The Concept & Script (45 minutes)

I started with a rough idea: What if grandparents tried to explain modern B2B marketing problems using Gen Z slang?

The contrast felt right. We get lost in buzzwords, but grandma gets ICP Marketing.

I used PJ Accetturo's approach from his Puppramin (spoof) and Kalshi (aired on TV) commercials as a role model. Here's my exact starter prompt to Claude:

Act as a viral video producer.
Give me 4 different scripts for a Keyplay video commercial that I'm gonna produce with AI and share on social media.
I am going to make the commercial using Veo 3. Research Veo 3 best practices so that the script can be easily translated into prompts.
The commercial should be no more than 6-8 total shots. Ideally ~30 seconds total. It needs a killer hook.
I want each shot to be a different person in a different situation so it's easy to produce each piece in Veo 3.
The ending bit should be a tag line where it's like "ICP solved" and then cuts to the Keyplay logo.
Approach to consider: Old people talking about B2B marketing. Nobody has 90-year old influencers. Maybe they use a bunch of Gen Z language like bro and W's in the chat and "cook" etc. Little statements about ICP and Keyplay. Then it could end with something like 9 out of 10 grandparents recommend Keyplay. Not sure on this idea just riffing.

The first script was mid.

I spent 30 minutes iterating with Claude. My team jumped in with feedback. Lauren had the key insight: make the grandparents read their lines deadpan, like they're just reading from cards. That's what made it funny.

The Final Script

After iterations, here's what we landed on:

Shot 1 - Grandma Reading Deadpan: Selfie video of 85-year-old grandma with thick glasses, holding camera at arm's length. She's clearly reading from something off-screen in a flat, confused voice: "Spray and pray marketing is... mid. These companies be spending millions targeting everyone. That's giving... broke energy?" She squints at the paper. "No cap."
Shot 2 - Grandpa Struggling: Selfie video of 90-year-old grandpa with oxygen tank, holding index cards. Reading slowly: "Sheesh... bro? Thirty percent of marketing budget wasted on non... I-C-P accounts? That's an... L." Looks at camera confused. "Martha, what's an L?"
Shot 3 - Street Interview: Street interview with elderly woman. Reporter asks: "What's your take on Account-Based Marketing?" She pulls out reading glasses and a crumpled paper: "A-B-M platforms? That's... chew-gy... A-F. Black box scoring has no... rizz?" She looks at reporter: "Is this English?"
Shot 4 - Distinguished Man, Zero Context: Selfie video of 80-year-old man in suit, reading from phone with arm extended: "Intent signals are... sus. These companies are... delulu... if they think predictive platforms... slap." Long pause. "What does slap mean? I thought that was bad."
Shot 5 - Chess Player's Take: Selfie video of elderly Black man at chess table. He holds a brochure that says "Modern GTM Strategy." Reading slowly: "They said... just trust the AI scoring model. That's like playing chess with no pieces." He looks at camera. Long pause. "You need a strategy. Not vibes."
Shot 6 - Eastern European Grandma: Sharp-tongued elderly woman at kitchen table covered in spreadsheets. She deadpans: "You target 10,000 accounts." Raises eyebrow. "You book four meetings." Beat. "My grandson could do better with a phone book and a pencil." Leans back: "Get your ICP together."
Shot 7 - The Resolution: Elderly man at computer desk. Screen glows on his face. He leans toward webcam and says softly: "ICP solved." Nods once. "Grandparent approved." Fade to black.
Shot 8: Keyplay logo + "ICP marketing. Grandparent approved." Note: Added this at the end using iMovie. Real editors use something like CapCut or Adobe Premiere, but you can ship something with consumer grade tools for sure.

Creating the Shots (2.5 hours)

This was the meat of the work toggling between Google Flow (Veo 3 model) and Claude. For each shot:

Ask Claude to write a detailed Veo 3 prompt for each shot, based on the script.
Paste into Google Flow and generate.
If it was off, back to Claude for tweaks.
If it was close, just regenerate (Veo 3 can be random).
When stuck, I asked ChatGPT for a second opinion.

The key was being specific about style, setting, and audio. Here are three examples with the full prompt. Each took iteration.

Shot 1 Prompt (Deadpan Grandma):

A selfie video of an elderly white woman in her mid-80s with thick tortoiseshell glasses and curly silver hair. She wears a lavender cardigan over a floral blouse. She holds the camera at arm's length, and her thin arm is clearly visible in the frame. The style is YouTube Vlog like. The character squints at a piece of paper in her other hand, occasionally glancing at the camera with confusion. Behind her is a cozy living room with doilies and family photos. She says in a flat monotone voice: Spray and pray marketing is mid. These companies be spending millions targeting everyone. That's giving broke energy. No cap. Sounds: The quiet ticking of a grandfather clock, the distant hum of a television in another room. No subtitles.

Shot 3 Prompt (The Street Interview):

A street interview style video. Over-the-shoulder shot of a young attractive asian female News reporter holding a microphone with a news logo interviewing an elderly jewish woman. An impeccably dressed elderly woman in her late 80s with perfectly coiffed silver hair faces the camera. She wears a Chanel-style bouclé jacket, massive diamond earrings, and designer sunglasses pushed up on her head. Background: A busy day in New York's Times Square. Reporter speaks first: What's your take on Account-Based Marketing. Elderly woman responds by reading from a small card in a refined accent: A-B-M platforms are NOT the answer. Elderly woman continues: Black box scoring has no rizz. Elderly woman stops reading and lowers the card. She looks at the camera, frustrated now. Sounds: Bustling city traffic, taxi horns, tourist chatter. No subtitles

Shot 4 (Distinguished Man, Zero Context)

A vlog show video of an elderly Indian man in his early 90s with wrinkled dark skin and thin white hair. He wears a crisp white button-down shirt with a navy cardigan and wire-rimmed glasses. The style is YouTube Vlog like. Behind him is an elegant study with leather-bound books and framed university degrees. He holds a smartphone in his other hand and reads from it with clear confusion: "Intent signals are sus. These companies are delulu if they think predictive platforms slap." He puts the phone down completely. Long pause. He looks directly at camera and asks in a refined voice: "What does slap mean. I thought that was bad." Sounds: A grandfather clock ticking, pages turning somewhere nearby. No subtitles.

The pattern: specific visual details, clear audio instructions, environmental sounds, explicit "no subtitles" directive.

→ Get all 7 prompts and see my iterations in this Google Doc 📃

What I Learned

This isn't one-shot magic. You work with the AI, not just prompt it. My first two shots took forever. By shot 7, I had a feel for what works.

Veo 3 is a slot machine. Same prompt, different results. Audio sometimes works perfectly, sometimes doesn't. Subtitles appear randomly. Just regenerate.

Simple beats complex. Single speakers work better than multiple. Clear, short dialogue beats long monologues.

Details matter. "YouTube Vlog like" got me the selfie angle I wanted. Specifying "no subtitles" usually worked. Environmental sounds added production value.

There are levels you can explore:

Level 1: Type one prompt, get an 8-second video. Smile, walk away. 90% of people stop here.
Level 2: What I did. String 6-10 shots into a story you can use. Now I have the muscle and can repeat when the right idea hits me.
Level 3: Push quality. Add ElevenLabs audio, better narrative structure, iterate more upfront, test with audiences. I might try this next if I get a good idea.
Level 4: Professional work like the Kalshi ad (hundreds of attempts + pro editing). If you’re already a great script writer and editor with taste, then this can be a force multiplier.

Cost: ~300 Veo credits. About $25 + the monthly cost of a pro plan.

Time: ~4 hours total. I set a hard deadline (picking up my son at 3:30pm) which forced me to ship v1 instead of endless tweaking.

Why This Matters for B2B Marketers

Most AI video examples online are B2C. Consumer brands making funny ads. Fashion shoots. Movie trailers.

But B2B marketing needs video too.

The tools are here. They're only getting better. And there's no playbook yet.

That's the opportunity.

We're all just figuring this out. There are no AI-Native marketing experts because nobody's been doing this long enough to be an expert. The best people are just trying stuff, finding leverage in surprising places, sharing what works.

This video won't win any awards. But it made people stop scrolling. It started conversations. It showed our positioning in a way that a blog post couldn't.

And the best reward: I now have a baseline understanding of AI video that I can build on. I know a little more about how it works, what it can and can't do, and have planted the seed for future ideas.

That's worth an afternoon of exploration.

Resources That Helped

Greg Isenberg interview with PJ Accetturo - Watch this first
Kalshi Video Breakdown - See the prompts
Veo 3 Prompt Advice - Technical tips
Example Veo 3 Prompts Collection
Google's Getting Started Guide