TL:DR – Some of my experiments with so-called AI for creating a series of interesting images with various themes and styles.
I have recently been using ChatGPT, the language model, to help me come up with different types of headlines for articles. I haven’t used any that it has suggested, but I have prompted it with the text of my articles and ask it to act as a sub-editor on a particular type of publication.
The headlines it devises are always interested, I usually ask it to tweak in a particular way, and usually then re-word it to my own taste. So, for this article, it suggested: “Exploring the Possibilities of AI-Generated Photorealistic Images” for that’s the content to follow this. I asked it to make the headline shorter and use a pun and it replied: “Painting a New Picture with AI-Generated Art”. That’s not bad. Another tweak: “AI Canvas for Next-Generation Photorealistic Painting”.
Its suggestions led me to my own headline: “Picturing new art with the AI paintbox”.
Okay. So, like I say, I’ve been experimenting with AI. A while back I did some Wombo Art, which you may have seen on here. Photorealistic output but surreal. The AI world has moved on apace with the likes of Dall-E (pronounced like Salvador’s surname), that generates weird images too. I’ve not delved into Stable Diffusion yet, but I did take another look at MidJourney this week, which functions via a server on Discord. I was inspired to try it by a popular social media update this that showed four portraits, purportedly of people painted during the renaissance, but who had more than a passing resemblance to the actors from the BBC comedy Blackadder. Indeed, they were created with prompts that mentioned Rowan Atkinson, Tony Robinson, Stephen Fry, and Hugh Laurie. They were very good.
I thought I’d like to create something not dissimilar but with a different twist. I prompted MidJourney to generate various well-known, classic portraits of women but where they are all wearing a covid facemask.
The first prompt I tried was to recreate Vermeer’s Girl with the Pearl Earring but wearing a mask, then Leonardo’s Mona Lisa, Botticelli’s Birth of Venus, Whistler’s Mother, Warhol’s Marilyn Monroe, and an AI self-portrait of Frida Kahlo (my personal favourites).
The thing to remember is that none of these were images anywhere in the world before I gave the AI a text prompt to trigger it to generate them. It has been “trained” on a large data set of images and “knows” about artists, their paintings, and many different aspects of culture and the world around us and draws on these [pardon the pun] to generate an entirely new image.
I also came up with various other prompts such as asking for a renaissance portrait of David Bowie. This was the best of several attempts
You can also prompt it with an actual image. So I did and asked it to create a cartoon caricature from my website photo. This AI triptych could almost be a portrait of me as a young man after I first cut my hair short, the middle one could’ve been me in my late 40s, and that last one is perhaps me in 20 years time!
I also fed MidJourney a photo of my band C5 performing live and asked it to generate a photorealistic cartoon image…not entirely sure what its “thought” processes were in this case.
There are endless possibilities, going back to the mask theme, I prompted the AI to create a picture of rockstar Peter Gabriel in a fantastical covid mask. The results, as reader Keith Walker suggested, are very Silence of the Lambs on Broadway.
For a serious article about assistive technology on smart phones for the partially sighted I prompted the AI to generate an appropriate image
I did so much messing around with MidJourney this week, that the software told me I’d used up all my free trial credits and I must now subscribe to carry on using it…