AI Image Generation with Midjourney and Dall-E 2

Like many creatives, I've spent the past few months experimenting with AI image generation programs. Beta trials began over the summer for Open AI's Dall-E 2 as well as Midjourney. These programs, as well as Stable Diffusion, Wombo's Dream, Google Imagen and Disco Diffusion are still in their early stages. Much has been discussed about AI image generation in the short time it's been around.

One critical thing to understand about AI-generated art - though it uses a vast array of online images as a source, these programs never borrow pixels or "photobash" to combine existing images. Instead, they take user-generated text prompts and compare them to images using those terms, and analyze the results to deeply understand the connections. The programs then synthesize the results to create completely new images based on the prompt.

Here's a quick exploration. I used Midjourney to make the images below, based on the prompt "magical land, rolling hills, rainbow river, sunbeams, puffy clouds, trees, 8k render, --w 9000 --h 6000" – the last two being parameters for width and height.

Midjourney returned these four low-res options:

I liked the upper right option above, so I asked it to give me variations of that one:

I preferred the upper right result from the options above, so I had Midjourney upscale it. This is a partial upscale:

I liked the results, so I first did both a light upscale from the image above –?which has minimal details added:

I also had it do a more detailed upres:

I then took the more detailed upres into a different AI art program, Dall-E 2 –?which, unlike Midjourney, only works in a square format (at least at the moment). Dall-E 2 lets you edit images, so I cleared out a space in the middle of the Midjourney render and asked Dall-E 2 to add "tall majestic castle photorealistic" in the empty space.

Dall-E 2 came back with three options, filling in the deleted space seamlessly:

I liked the first result above, so I asked Dall-E 2 to create variations:

I still preferred the original image, so I stuck with it:

These programs are moving forward at an incredible rate – the original Dall-E came out in 2021 and was much more primitive than Dall-E 2. Midjourney has already released three updates to its algorithms since launching in July. While the results are mindblowing, the programs that generate them are still in their embryonic stages. Heavy debates have already started taking place amongst designers, illustrators, 3D artists, and other creatives around how AI art programs will affect the creative arts in the future.

If you're using these programs, leave a comment with your thoughts about your explorations so far.