Experimenting with DALL-E Image Generation
I got access to DALL-E the other day, which, of course, meant that I had to experiment with it straight away.
I did start a Twitter thread showing all of my generated images, but I thought it would also be good to post them here.
So after using my first 20 credits, here's what I came up with.
You will need to click on each image to see the full-size version. (I didn't want to auto-load >100MB)
First of all, I thought I'd start with something fun:
"A grassy landscape showing a Pikachu in its natural habitat
I got four images as usual, but I also decided to generate three more variations to see what that was like.
After that, I was thinking about creating some futuristic cities with trains (I'm a big fan). It didn't go how I planned but still went okay.
"A realistic photo of a planet where trains are the only mode of transport"
Then there were a few more revisions as I tried to figure out how to get what I wanted.
"A realistic photo of a city where trains are the only mode of transport and nature is abundant"
"A realistic photo of a futuristic city where trains are the only mode of transport and nature is abundant"
After a few changes, I finally came up with something close to what I had in mind.
"A futuristic city where trains are the only mode of transport and nature is abundant, digital art"
Once I had a bit of experience, I decided to go back to Pikachu.
"A macro photo of a realistic pikachu in a realistic landscape with a shallow depth of field, digital art"
Kevin Wammer asked if it knew about the Steam Deck, so I tried two descriptions out. I'm not sure they really worked.
"an abstract painting of a steam deck, digital art"
"an synthwave style image of the steam deck games console, digital art"
Then skateboarding came to mind.
"an abstract painting in a synthwave style of a skateboarder doing a kickflip over a set of stairs in a lush futuristic city, digital art "
"A 3D render of a skateboarder doing a kickflip over a set of stairs in a lush futuristic city, digital art "
Of course, I had to then add in Pikachu.
"A photo of pikachu doing a kickflip on a skateboard in london"
After that point, it really was a bit random. I was thinking about green cities, Pokemon, and cats, and I think you'll find the odd one out that came from my girlfriend (bubble tea).
"A photo of a futuristic city full of nature where Pokémon roam free, digital art"
"a photo of london in the year 3000 except there are no cars and it is full of nature, digital art"
"a photo of a cat on a boat drinking bubble tea"
"a photo of a cat travelling to a lost city on a boat going through a river in a jungle, digital art"
"an oil painting of a cat travelling to a lost city on a boat going through a river in a jungle during golden hour, digital art"
"an oil painting of a group of animals resting by an open fire in a lost city in a jungle with a river, digital art"
"a photo of a group of animals resting by an open fire in a lost city in a jungle with a river, digital art"
"A Charmander sat on a tree stump in the middle of an open woodland at golden hour, digital art"
I think it's clear the images were getting better as I started to work out the type of language needed. I prefer the last few scenes myself, and I was surprised to see how they turned out. I'm definitely going to be playing with this again soon, but maybe after I've read a bit more about it, and try to come up with more advanced generations.
A new AI system that can create realistic images and art from a description in natural language ↩︎