(this will probably truncate in Gmail - sorry! Click through though, it’s a fun little story)
A few months ago, I got access to DALL-E, a program from Open AI that converts plain language descriptions into images. I decided to have a little fun with it, so I used it to illustrate a story that I used to tell my kids when they were small, about a small mouse named Pic McPail, who was constantly getting in trouble by building machines to do his chores (Pic is lazy). The machines never work right.
What follows is the result of about 20 minutes of working with DALL-E to generate illustrations for that story. The words are all me, but the images are all DALL-E.
Some interesting learnings from this. The first is, at least for this problem, DALL-E did a really great job of making images that worked - I think it's no exaggeration to say that this is the first time in maybe 10 years that I've had this kind of jump in personal capability from a piece of tech (maybe mobile maps before it had this degree of impact).
To a large degree, I got what I wanted within one or two tries. The most challenging image was the machine breaking - it's such a weird idea that that I don't there was much similar in the training set. But it did work!
And, interestingly, one of the other images that I struggled to generate, the oatmeal breaking out of the house and flowing down the hill, resulted in the "surfing" image, which I thought was so cool that I changed the story slightly so I could use it. So that's an interesting example of the machine collaborating and generating a new idea!
The final challenge is that DALL-E doesn't really have a sense of "continuity". So, even though I always started the prompts with "a watercolor painting of a mouse with a blue cap...", it didn't really generate images that are all in the same artistic style, or make the mouse look (sort of) the same each time. The continuity you see is because I curated the results to look similar.
All of these, to me, point in really interesting directions. If I expected to just give DALL-E the text of the story and say "please illustrate this" and then expect it to get it right on one try, I'd have been very disappointed. But since I used DALL-E as an iterative tool and a partner, and was able to apply my own refinements and curation, and even take accidental "suggestions" from it, it worked incredibly well as a creative tool - it produced a result in a very small amount of time, that I would have been completely unable to produce, in any amount of time (I’m a bad artist). This seems like a good model for using emerging AI capabilities in general - not as a drop-in replacement, but as an enhancement for what people do already, and sometimes an enabler of entirely new capabilities in combination.
On to the story:
Pic McPail was a little mouse who lived in a little house on a hill. He was generally a very good little mouse but sometimes, he got himself in trouble.
One day, Pic didn’t feel like making his oatmeal for breakfast. “I hate making oatmeal”, he thought, staring at his bowl. “I’ll make a machine to make it for me!”
So Pic went down to his shop and got some tools out of his toolbox. He got some metal, and some wood, a few rubber bands, and other bits and pieces he’d found around town.
He started working on the machine. It didn’t look bad!
He turned the machine on! It worked….but it worked a little too well!
It started running, and running…Pic ran over to turn it off, but he had forgotten to put in an off switch! Oatmeal was pouring out of the machine!
More and more! The machine went crazy! It broke! Oatmeal was going everywhere!
Pic ran out of the house. Oatmeal came pouring out of the doors and started running down the hill.
He was about to be covered in it! Fortunately he grabbed his surfboard at the last minute. He surfed in the milk and the oatmeal, down the hill.
But the oatmeal was heading for the village! They were going to be so mad at him! They just finished cleaning up from the whipped cream balloon disaster (which would have totally worked except for those pesky bees!), and now he was going to mess it up again.
The village was buried! Pic crashed into the town hall. All the villagers came out to yell at him!
But Pic’s friend Mrs. Willey scooped up some oatmeal and made cookies.
She had a big tray that she took out to everyone. That helped calm them down.
Everyone had a cookie, and settled in to clean up.
Pic was happy. He didn’t have to make breakfast, plus he had a cookie instead of oatmeal! After he cleaned up the village, the hill, and his house, and got his surfboard off the top of the church, he was so tired! He went home and took a nap.
His couch wasn’t too sticky.
Add children's author to the resume :)
But here – it does! It creates the "continuity" feeling, as you can see in these series:
https://habr.com/ru/company/ruvds/blog/678876/
So, I guess, you might just not know how to do this!)
> The final challenge is that DALL-E doesn't really have a sense of "continuity"