I Tried AI to Make a Personalized Cartoon. The Results Were Comical

It started as most adventures do: without much thought as to where the twists and turns would take me. I set out on the heels of the unknown for six months, or for as long as it "felt right." That was eight years ago, and I ended up more than 10,000 miles away from home. 

Now artificial intelligence is giving me a creative way to keep in touch with family on the other side of the world, and to give them a glimpse into my life away from home. 

After three years hopscotching continents, I landed in New York City, enamored by its energy. Writer falls in love with New York City -- how original. 

While my family and friends were thrilled I finally found my city soulmate, grief soon ensued. It was no short road trip or flight to see loved ones. 

I'm good about going home once a year and calling family and close friends regularly, but the timezone makes spontaneous chats challenging. After a long day of work, the last thing I want to do is get on a FaceTime call at 8:00 p.m., no matter how cute my nephews are or how much I miss the Aussie banter. 

I found myself doing more audio calls because it's easier on my energy. With video calls as close as I can get to replicating in-person experiences, I was conscious of not compromising my relationships. I needed some creative ways to connect long-distance, also in new async ways. 

I thought, wouldn't it be cool to make a mini movie for my nephews, using AI? I was tossing up between Midjourney or Dall-E, but decided to go with Dall-E because the new version is available through ChatGPT Plus.

Dall-E 3 is the latest text-to-image model by OpenAI, the creator of ChatGPT. It was first released in January 2021 and is in its third iteration. 

Getting set up 

To make a movie, I needed: 

1. The script: ChatGPT to create the narrative and script. 
2. Illustrations: Dall-E 3 to generate images. 
3. Animations: A free image-to-animations tool. 
4. Voiceover: A text-to-speech voiceover tool. 

Movie making in Dall-E 

What's all the rage with nine-, eight- and five-year-olds? Bluey. I wanted to see if I could create some kind of mini episode inspired by Bluey, starring my nephews and me. 

Step 1: The script 

Prompt: "Make an adventure story about Bluey, Auntie Amanda and her life in America. Focus the narrative on how she misses her nephews, Lucas, Zach and Jackson, but is on a big adventure in New York City. It should feel like a mini Bluey episode that's entertaining and educational. Include visual descriptions for illustrations and text to be narrated."

Here are the first two scenes of the seven-scene script it came up with:

Not bad at all! My nephews have written to their teachers that they can't wait to visit me in America and see all the tall buildings, bridges and people, so I asked ChatGPT to add that in. 

It was pretty close to something I liked already, but I made some small tweaks to add in more personality and relevance. There was one line that sounded like I had passed away, so I changed it. 

Step 2: The visuals

Next, I copied all the illustration prompts into one document and pasted it in ChatGPT. I started by putting all prompts in at once, but later decided to do one at a time so I could give feedback, like changing the hair color to blonde. And when I inputted it all at once, the characters lacked consistency -- one image had short hair, and the next had long hair and bangs. 

Prompt: "Please create an animated illustration in the theme of a Bluey episode: Insert illustration prompt." 

It took a few attempts to understand what I was looking for. It kept adding in characters I didn't ask for. 

For example, I liked one frame but had to ask it to add Bluey into the picture. But then our family suddenly became mixed-race, and we went from three to two nephews. 

Dall-E continued in a daze. I asked it to use the characters in the right-hand photo for the rest of the visuals. But then my nephews suddenly had brown hair... and were teenagers.  

Worried about the lack of consistency between illustrations, I went back to putting it all in one prompt but prefacing the physical features. 

No such luck. In one sequence, I went from a chic middle-aged woman, to a Black teen, then an elderly woman named "Aunting Bongo."

At this stage, I was ready to throw in the towel and go back to good ol' FaceTime. 

I went back to the first lot of images, which ironically were the best, besides the bangs. And Dall-E turning Bluey into a babe.

I saved each image and then uploaded the ones I wanted to edit, with feedback. This strategy seemed to work, but I had to settle for some inconsistencies, because Dall-E kept backtracking -- or turning me into an elderly woman with six nephews.

It was comical at this point, so I surrendered to it being imperfect. 

Step 3: The animations

I needed to turn these visuals into videos. I thought I could do it all in Dall-E, but I had to use an external free tool, Pika. For the sake of simplicity, I went with slight movement animations, nothing too theatrical.

Step 4: The voiceover

Next, I copied all the narration text into one document and pasted it in the free AI voiceover tool, ElevenLabs. I picked a voice I liked, then generated and downloaded the voiceover. The script was too long, so I had to shorten it. 

Putting it all together 

Over on Vimeo, I uploaded all the animations and voiceover -- you can watch the final product here. It wasn't perfect, and it took a few hours to create it from start to finish, but it was a fun little digital gift I could surprise them with to remind them that I miss them. 

Or maybe it'll make them laugh. Both instances are a win. 

But I won't be giving up FaceTime anytime soon.