Jamena McInteer
Posted on December 16, 2022
The past few days I have been playing around with DALL·E and testing its capabilities, and seeing what I can do with it. DALL·E generates 1024x1024 images based on a text prompt, in the style described by the prompt. With the editing feature, it is possible to extend the picture and edit parts that aren't quite right. Using these features, I set out to create a couple of compositions.
Abstract Painting
I decided I would create an abstract painting with teal and orange colors as I love bright abstract art. I entered the following prompt:
“An abstract watercolor painting with splashes of teal and orange and highlights of glittery gold gouache paint”
I thought these were all amazing, but particularly liked the third one. I then overlapped a generation frame with part of the painting and used the same prompt to extend the painting.
I could have continued to make it larger and larger, and I may do so in the future. The AI seems to do especially well with abstract art, which makes sense as there are few rules when it comes to abstract art.
A photorealism oil painting of chickens
Next I wondered what it could do with something less abstract. I entered the following prompt:
“A rooster and hens, photorealism oil painting”
Except for the last one, I quite liked them. However, the middle hen in the first one had a weird head and the rooster had a kind of creepy hollow-looking eye. I decided to expand on the second one.
I couldn't reuse the same prompt, as I would have ended up with a lot of roosters. Instead I had to describe the section of the painting I wanted, while overlapping the generation frame with the existing painting so the style would carry and the edges would blend. I found that the further I got from the original image, the more the style drifted.
“One hen”
It also struggled a bit with sizing and perspective, depending on how much the existing painting was overlapped. On of my frames, with the following prompt, resulted in a large chicken that awkwardly dwarfed the existing chickens.
Some of the style drift was quite dramatic.
“Bush with some flowers and chicks”
“Hen and chicks”
A landscape country scene
Next I decided to create a country scene in the style of Martin Johnson Heade, an American landscape painting from the 1800s. I started with the following prompt:
“"A bright and surreal country scene with a red barn" by Martin Johnson Heade”
I thought this was a lovely start, and decided to expand on the first image. I started expanding using the same prompt, but then began editing. One of the expansions included a hint of someone sitting under the tree, so I decided to go with that and see how I could improve it.
I decided I wanted a little girl drawing under the tree, so I used the eraser tool to remove some of the elements in the scene that were under the tree, and then used a generation frame over the erased section and used a prompt for what I wanted to fill the area. I tried a few different prompts, but finally landed a generation from this prompt:
“"A little girl sitting and drawing in a notebook" by Martin Johnson Heade”
On the right side of the painting I wanted to add some horses. The AI seemed to struggle to paint good-looking horses, and sometimes they appeared in front of the fence.
“"Horses" by Martin Johnson Heade”
As in the previous painting, it also struggled with size and perspective.
I finally got a decent horse and decided to leave it as is.
“"A bright and surreal country scene with horses" by Martin Johnson Heade”
A weird, unrecognizable artifact was added on the left hand side next to the barn, so I used the eraser tool again and generated a windmill.
“"A bright and surreal country scene with a windmill" by Martin Johnson Heade”
I kept expanding the scene, using some of the following prompts to paint the sky:
“"A bright and surreal cloudy sky" by Martin Johnson Heade”
“"A bright and surreal sky" by Martin Johnson Heade”
“"A bright and surreal blue sky with small birds flying in the distance" by Martin Johnson Heade”
Again, the style started to drift the further I got from the original image. I tried to overlap about 75% of the generation frame with the existing image to minimize this.
The piece I ended up with was this:
There are small edits and improvements I would like to make to perfect it, but I thought it came out quite nicely.
Llamas!
For fun, I decided to generate some llamas in different styles to see what DALL·E would do with the same basic prompt but different style parts of the prompt.
“A photo of a llama wearing a teal and orange inca scarf and sunglasses”
“A llama wearing a teal and orange inca scarf and sunglasses, digital art”
“A llama wearing a teal and orange inca scarf and sunglasses, watercolor gouache”
“A llama wearing a teal and orange inca scarf and sunglasses, watercolor”
“A llama wearing a teal and orange inca scarf and sunglasses, photorealism oil painting”
“"A llama wearing a teal and orange inca scarf and sunglasses" by Rosa Bonheur”
I don't think the AI knows Rosa Bonheur, or maybe this is the wrong subject for that artist.
“"A llama wearing a teal and orange inca scarf and sunglasses" by Albrecht Dürer”
Same with Albrecht Dürer.
“"A llama wearing a teal and orange inca scarf and sunglasses" by Rembrandt”
I think the subject of the prompt makes a big difference on how closely the image matches the style of a particular artist.
“"A llama wearing a teal and orange inca scarf and sunglasses" by Johannes Vermeer”
Conclusion
Experimenting with DALL·E can be a highly enjoyable and creative experience. The tool's ability to quickly generate unique ideas allows for endless possibilities for exploration and inspiration.
However, working with DALL·E can also present challenges, such as quirks and weirdness to work through in order to produce something that looks good. Despite these challenges, the reward of successfully utilizing DALL·E to bring an idea to life can make the process well worth the effort. Playing with DALL·E can be a rewarding and enjoyable experience for anyone interested in exploring the capabilities of artificial intelligence and creativity. And now I've got to go buy some more credits! 😊
Posted on December 16, 2022
Join Our Newsletter. No Spam, Only the good stuff.
Sign up to receive the latest update from our blog.
Related
November 29, 2024
November 13, 2024
October 23, 2024