OPUS 500

 kw: art generation, ai art, simulated intelligence, milestones

I have been learning how to use art generation programs since November 2022. That is just a little more than two years. Initially DALL•E2 was all that was available. I used it until it went out of service a year later in favor of DALL•E3, which powers the Bing Image Creator. In the first few days using DALL•E2, with lots of experimentation, I produced two images that still please me a lot. One is a still-life "painting" in the style of Cezanne, the other is a "painting" of mountains in the style of the Hudson River School of painters:

Note the color blocks at lower right. That is DALL•E2's "signature". Pictures by DALL•E2 were 1024×1024, and to make them bigger one could Outpaint. I did that a lot. Eventually I began to make images that would be useful as screen wallpaper. I would prepare a large image and then crop it to 1920×1080, or a larger size with a 16×9 aspect ratio, and sign it in one of the lower corners with my name plus "/ Dall-E2", to give credit to both myself and my "commissioned artist".

At the very beginning I also tried out MidJourney and Stable Diffusion, but I wasn't too thrilled by the available toolsets, and SD required Discord, which I don't like to use.

Fast forward a couple of years. Over time I learned of other art generators:

• PlaygroundAI - I used it for 2-3 weeks during January 2024, but soon came across Playground, which may be related but had a better credit and subscription structure.
• Playground - I loved the Canvas mode, which also had Outpainting. It was taken out of service in November 2024, the company having decided to concentrate on graphic arts rather than fine art.
• I had been using Bard in Chrome as an alternative to ChatGPT. When its name was changed to Gemini, it was announced that it could generate art also. At first it would make four images at a time, but then it dropped three and now it only makes one per prompt, and it is always square, but 2048×2048. I've used it since February 2024.
• I found Leonardo AI in August 2024 and I've been using it since then. After a few months of using the free service, I bought a passel of credits.
• DreamStudio showed up the most recently, in November 2024. After a couple of months testing the free service, I bought some credits.
• Lastly, also in November 2024, I found Google Labs' ImageFX, which is free, but they don't keep history. Download what you want to keep right away or it'll vanish. It has some interesting features that I like, but not the rich feature set of DS or Leo AI.

My current "stable" of artists is DALL•E3, DreamStudio, Gemini, ImageFX, and Leonardo AI. All of these except Gemini can make images with various aspect ratios, although in the case of DALL•E3 the only alternative to square is "Resize to 4:3" which is actually nearly 16:9 (1792×1024), and it is a fresh image, similar to the one you "resized", but not a direct expansion of it. 

Recently I checked the folders that contain my "wallpaper" images, and found there were nearly 500. I had decided to try a different kind of image, similar to a piece of art we have that is embroidery on a floral theme. I wrote a few prompts and tried them out with the various art generators. I concluded that Leonardo AI and DALL•E3 worked with these prompts the best. At first I wanted a depiction of one artwork in a horizontal format, but none of the programs offered pictures that filled a 16×9 frame, or nearly so. Leonardo AI prepared a couple of images that featured two pieces side-by-side on a wall, so I began to concentrate on that. This is the prompt that produced the pictures of lilies:

Two pictures, finely detailed embroidery, of lily flowers of various colors and foliage within a pale green mat and dark green frame, mounted next to one another on a white wall

The upper image is of peonies, with a similar prompt but without "various colors". I also wanted an image with different flowers in each frame, and tried a prompt that said something like "peonies" first and "roses" next, but I kept getting peonies everywhere. Then I tried "roses" followed by "peonies", and still got only peonies! It seems that in existing art, peonies are a much more common theme for embroidery than roses. So I wrote a prompt that only mentioned roses, and got some images showing two pieces of art with roses. As I went along I produced wallpaper images from the ones I liked best.

I noticed at this point that I had 499 wallpaper files. I fired up Gimp (a Photoshop "cousin"), copied roses from one image with all roses and pasted them into one with all peonies, to produce my 500th wallpaper:

PS: I know that "opus" mainly refers to musical works, but I have seen it used for other creative products also.