The Thanksgiving Day War!
Hey everyone, It's me. I know I'm fashionably late to the party—a couple of days behind for the folks in the U.S., but hey, just over ten months ahead for us Canadians, right? But let's not get bogged down in the details. Many of you know that I've been a little obsessed with AI lately. In that vein, I've been tinkering with something pretty cool (okay, maybe more on the cringe side than cool) and I'd like to share it with you. It's a quirky little project I'm calling "The Thanksgiving Day War," and it's a mash-up of some cutting-edge AI tech: GPT-4, DALL-E 3, and Descript. The idea? Use AI to do (almost everything) which, as it turns out, is harder than it sounds since it involves a lot of steps.
Let's kick off with ChatGPT (Plus) and the GPT-4 model. A few days ago, as the American Thanksgiving (as opposed to plain old Thanksgiving) was approaching, I saw a cartoon that had a turkey trying to avoid the axe, and it made me think, "What if the turkeys had human level intelligence, access to weapons, and decided to go to war with humanity?" The light bulb went off. I opened up GPT-4 and asked it to write a short YouTube video script about a revolt by turkeys, with help from Canada geese, mediated by a rooster. No, I don't know what made me think of this, but ChatGPT happily spit out a script for me.
It doesn't judge.
Seconds later, with script in hand, I realized that I needed visuals. Enter DALL-E 3, GPT-4's artsy sibling, also from OpenAI. They say that "a picture is worth a thousand words" and the reverse is what we call a prompt in generative AI parlance. I took each of the paragraphs and fed it into DALL-E, asking it to "create a whimsical portrait orientation image inspired by" and then fed it the text from each paragraph. While I admit that I did not just take the first thing the AI spit out, it didn't take too long before I had something I was, more or less, happy to work with.
The last piece here is Descript, an amazing but somewhat unusual video editor, popular with creators of various stripes including podcasters. As video editors go, it's a little on the weird side, but in a good way. You edit videos using an interface that's akin to a word processor. If you cut out a piece of text, it takes the accompanying video along with it. Honestly, it takes some getting used to, but the results are worth it. It has all sorts of really cool features that are quite impressive. For instance, it lets you clean up your voice and get rid of background sound, and it can make sure your eyes keep looking at the camera, even if you're reading from a script below said camera. It's pretty wild.
But here's the kicker—it can also clone voices. So, the voice narrating the video? That's me, or rather, it's the Descript's version of me, AI Marcel. It's a bit surreal hearing your own voice being generated from the script, and it's not quite perfect, but it's not bad either. When you get to the end of my post, check it out and decide for yourself. I did all my editing in Descript, breaking the text here and there to change the DALL-E images, adding some sound effects, and putting in background music. It's not just, "ask the AI," and everything works, not for this part anyhow.
So, that's the scoop on "The Thanksgiving Day War," a little adventure in AI-assisted storytelling, imagery, and voice production. It's a bit out there, a bit experimental, but hopefully that's what makes it fun, right? Whether it tickles your fancy or leaves you scratching your head, I'd love for you to check out the video and give me your feedback.
Until next time... Like. Share. Subscribe. All that stuff!