Morten Just

An app that makes tv ads for to one single person – using GPT-3, stock video, stock music, and a text-to-speech API

There was a new Nike ad last week. One thing I noticed was how it seems to follow a pattern; music, video clips, an emotional script talking about the big things in life.

What happens if you show GPT-3 a few Nike ads and ask it to create a similar ad for another company? Or maybe for one, specific person?

I decided to find out, and began hiring my robot production crew:

• Script and clips curation: GPT-3
• Voice: Replica Studios
• Free video footage: Pexels
• Free music: Unminus

With the robots on board, my job is to provide a company name and a product that company names. It then uses the three Nike examples to write a script and choose some videos.

In 2003, it took me a full hour to get ready for a 20 minutes run, and not (just) because I lacked the motivation. Wearing my full running outfit, I’d record a streamed radio show, convert it to MP3, and put it on my 1st gen click-wheel iPod. I had to hack it. It’s easier today, obviously. But the actual listening experience hasn’t changed much apart from the “skip 15 seconds” button. Podcast apps today still feel like hacked music players.

Discovering a new podcast isn’t much better either. If you think it’s hard to pick a movie on Netflix, try picking a podcast. Itunes now counts more than 1,000,000 shows – twice as many as just two short years ago.

Sign up for the private beta test at the end of this post

Yet, our podcast players look like they did five and ten years ago, when you subscribed to the only ten good shows and that was it. Today, shows worth listening to run into the thousands. Subscribing to all of them would make no sense. Why can’t we subscribe to topics? Or the people who are being interviewed? Or all currently trending podcasts about books?

That was the first thing I built. Let’s take a look, starting with discovery and subscriptions. Or rather, no subscriptions.

 

Read more on Medium