Back to home

Reinventing TV ad production with generative AI

Designer · Generative AI, Video, Post-production

Client ads and results are confidential. The ad and figures shown were recreated for this case study using the same process.

Problem

Streaming TV let local home service businesses target their own neighbourhoods, but a traditional 30-second spot still meant a crew, a location, talent and weeks of production, at a cost most local businesses couldn’t justify.

Outcome

We built TV ads almost entirely with generative AI, at a fraction of the cost of a traditional shoot. Local businesses got on the big screen at an affordable rate, streaming on NBC, Roku, DIRECTV and FOX, and saw a clear lift in branded searches and engagement.

From storyboard to production-level quality

From the voice to every scene, we made each ad with AI, then cast, prompted, cut and finished it for broadcast, so a local business could hold its own next to national brands.

From storyboard to production-level quality

From the voice to every scene, we made each ad with AI, then cast, prompted, cut and finished it for broadcast, so a local business could hold its own next to national brands.

From storyboard to production-level quality

From the voice to every scene, we made each ad with AI, then cast, prompted, cut and finished it for broadcast, so a local business could hold its own next to national brands.

Inside the production

Thirty seconds leaves no room for filler. Every ad moved through the same four stages, with our producer and designers directing AI at each one. The goal never changed: make a local business look established, sound like itself and hold attention until the viewer reaches out.

Inside the production

Thirty seconds leaves no room for filler. Every ad moved through the same four stages, with our producer and designers directing AI at each one. The goal never changed: make a local business look established, sound like itself and hold attention until the viewer reaches out.

Inside the production

Thirty seconds leaves no room for filler. Every ad moved through the same four stages, with our producer and designers directing AI at each one. The goal never changed: make a local business look established, sound like itself and hold attention until the viewer reaches out.

Strategy before a single frame

Concept, storyboard and script started with strategy, with AI helping where useful: who we’re reaching, what earns their trust and what gets them to call. Every shot was planned before it became a prompt.

Finding the voice of the brand

I generated every voiceover with AI. Firm and confident, or warm and family-friendly? I cast each voice by brand tone, audience and region, then directed the read until it sounded right.

Scenes that feel like the neighbourhood

I generated every scene in Google Flow with Veo 3.1, prompted from the client’s own technicians and trucks, in streets and homes like the viewer’s own. Most took several takes to feel local, never stock.

Cut to hold attention until the call

In After Effects, I brought the scenes and voice together with transitions, animated callouts and pop-up offers, pacing every cut to hold attention for all 30 seconds and land on the call to action.

Where the creative meets the screen

TV isn’t a bigger phone. There’s no click, viewers watch from across the room, and the ad is easy to ignore. Every design choice answered that: what they’ll remember, what they can read from the couch and what makes them pick up the phone.

Where the creative meets the screen

TV isn’t a bigger phone. There’s no click, viewers watch from across the room, and the ad is easy to ignore. Every design choice answered that: what they’ll remember, what they can read from the couch and what makes them pick up the phone.

Where the creative meets the screen

TV isn’t a bigger phone. There’s no click, viewers watch from across the room, and the ad is easy to ignore. Every design choice answered that: what they’ll remember, what they can read from the couch and what makes them pick up the phone.

Targeted by zip, versioned by zip

The account team mapped each client’s top-earning neighbourhoods. Every scene was generated, so each local version needed new prompts and a local offer, not a reshoot.

Built for the search that follows

Viewers look a business up after seeing it on TV, so every ad carried a search moment that made the name easy to remember and even easier to find.

Designed to be read from the couch

Viewers watch from across the room, so type was set large, the offer and search prompt stayed on screen long enough to read, and the call to action always stood out.

Learnings

Directing AI is a craft of its own. I learned to brief it like a film crew and to design shots around what the model does well. Veo struggled with people doing hands-on work, so a plumber fixing a pipe became a plumber inspecting it: same story, far more believable. Next, I’d test platforms like Higgsfield and build a library of consistent people, vehicles and locations, so every ad starts from a system of assets instead of a blank prompt.

Learnings

Directing AI is a craft of its own. I learned to brief it like a film crew and to design shots around what the model does well. Veo struggled with people doing hands-on work, so a plumber fixing a pipe became a plumber inspecting it: same story, far more believable. Next, I’d test platforms like Higgsfield and build a library of consistent people, vehicles and locations, so every ad starts from a system of assets instead of a blank prompt.

Learnings

Directing AI is a craft of its own. I learned to brief it like a film crew and to design shots around what the model does well. Veo struggled with people doing hands-on work, so a plumber fixing a pipe became a plumber inspecting it: same story, far more believable. Next, I’d test platforms like Higgsfield and build a library of consistent people, vehicles and locations, so every ad starts from a system of assets instead of a blank prompt.

Making stablecoin payments feel like everyday banking

Founding Product Designer · Product, Design Systems

Making stablecoin payments feel like everyday banking

Founding Product Designer · Product, Design Systems