Storytelling
ML/AI
Create magical content experiences with AI
Alex
•
Jan 5, 2023
AI-generated content is being used in tons of cool ways. Text-based models like GPT-3 are easy and accessible to use. People are generating poems in styles of poets like Rumi - building chatbots to tell Santa what they want for Christmas.
A month and half ago I was an absolute newbie to AI, and frankly I had no interest. I was working on a startup in Web2 and Web3 was stealing most people’s attention.
Yet each day my twitter feed seemed to be flooded with new demos showing how AI would slowly take over every medium of content that we consume.
I watched as AI content evolved from the classics with GPT3 like Jasper.ai and Copy.ai, to the amazing art creations on lexica.art, to full podcasts like podcast.ai, and even to my tiktok for you page — and enough was enough.
I was ready to get caught up.
Taking the best from each medium
So one day, I sat down with the smartest AI people at Founders Inc and dived right in.
From text generation with GPT3, to speech-to-text with Whisper, and of course image generation with Stable Diffusion, I wanted to figure out which mediums had the most unexplored potential.
Then, someone at the lab showed me something magical.
Before any of the AI avatar generators had come out, they talked me through this reddit post on generating Disney versions of yourself: https://www.reddit.com/r/StableDiffusion/comments/yf3ltx/new_dreambooth_model_modern_disney_now_available/.
I was blown away. Immediately, I thought my 8 year old sister would go crazy if I made her one — and soon enough, I facetimed her to display her very own Disney avatar.
Her reaction was priceless. In fact, she even showed her friends and they asked for one too!
It made me think back to when I was little and my next door neighbor had gotten me a custom bed-time story.
Soon enough, I was combining the Disney images with a story generated from GPT3, and then Eric from Unreal Speech showed me how to link in text to speech and have the stories narrated to you, and boom — MagicStory was generating AI content in 3 different mediums.
A few days later started building a way for more of my sisters friends to try it.
Bringing MagicStory to life
After the magic moment that started it, MagicStory felt like a fitting name.
I hacked on it through Thanksgiving and we launched this past week and have had hundreds of stories generated which was pretty cool :)
With the power of AI, MagicStory gives you a mini Disney studio team in your pocket, and it’s been pretty amazing to see the diversity of people that have made stories.
From my little sister and her friends, to friends my age, fellow founders, my old basketball coach, and even to Gary Vee! Check out some of the stories before to get a better idea!
And we have lots of improvements on the way! MagicStory for iOS, faster stories, stories tailored for different age groups, text-to-speech to read you the stories and more!
And if you wanna create your own MagicStory give us a try @ www.magicstory.ai :)
Alright, so how did I do this?
How does it work for the user?
Pretty straight forward actually. I have a form that you fill out when you click on 'Create a MagicStory' button. The form is super basic:
I ask for an email to send the story
Ask them what's the name of the hero
Describe the hero a bit (see image below)
Describe the setting (see image below)
Describe what happens in the story
Give a title!
That's it. Really.
With the power of AI, MagicStory gives you Disney’s studio team in your pocket. The main motivation for me to create this was to help kids learn to read faster and see themselves as anyone they can imagine.
This knowledge can be extended in many ways and for various other experiences. For example, buildspace used AI to create their first comic.
Check it out!
Tech stack I used to build this
Nextjs → the best way to build a web app
GPT3 davinci-003 → writes the story AND the image prompts
We send your story prompt to GPT3 (check out Buildspace’s GPT3 project below to learn to do it yourself!)
Next step is to fine-tune GPT3 with pre-existing good image prompts to have AI generate the new image prompts for us.
Finally, we use Stable Diffusion + Dreambooth to generate the images. We train a Stable Diffusion model on your hero to generate images of your hero in the story!
If you're confused, check out the project and note below. I literally learned everything from it lol.
Just start getting your hands dirty
This space is moving so fast. It's easy to feel paralyzed, but I highly recommend just starting with whatever you know. Webflow, Framer, Bubble (if you don't code), play with OpenAI playground, start testing out the API.
Ultimately, just ship whatever in a weekend. Sandbox your time and you'll be shocked at the progress you're able to make in just a few days! This will unlock further opportunities for you and you'll be on your way to becoming an AI master!
Hope you found this helpful! Please share whatever you're building in public and I'll leave my feedback :)
Good luck!
Alex
PS: also had the chance to meet Gary Vee!
