unslopping

I’ve released dozens of albums since 2018, and made all my own album covers except for one.

In some of those cases, I used generative AI tools. My attitude was different back then, and I’ll neither justify nor apologize for that now. But I’ve since felt that some of those covers are an embarrassment — I just didn’t know whether I wanted to go back and redo them, or merely acknowledge that I’ve moved on.

Second, I’ve lately been looking at some anti-AI art, and a common sentiment is that “bad”, amateurish human art is still better than genAI. And I get it. It’s a question of heart and soul and imagination.

Third, I’ve been exploring the possibility of using some other distribution service than Bandcamp in the future. Some of the options have a total ban on genAI, for the cover art as well as the music. I’d need to update the art for some of those older albums if I were to include those.

And fourth, reading Free Play: Improvisation in Life and Art has me thinking about art, play, the process, the joy of making stuff, and why humans make art.

So with that in mind, I’m redoing those album covers. At first I was just going to update the ones that look like AI art, which are especially embarrassing. But especially on the second and fourth points, I’m likely to do all of the ones that have a truly generative component.

Diversion is the first to be replaced, and here’s the new one:

The original one looked like AI, like a style that I don’t think any human would actually paint. (Part of why I originally thought it was okay.) It was a wonky suggestion of a wall of modular synth gear, and in the center either a dark rift sprouting tentacles/cables — or a figure in a long dark cloak and totally fabulous pink hair facing the synth, or perhaps both at the same time. I have always liked the ambiguity. But I’ve long since decided I disliked the style/appearance of the details, and the fact that this is something that a human artist should have made.

So I painted over it. Very crudely, as befits both my lack of that skill and the impression and process I wanted to go far. I thought about “computer art”, specifically a couple of electronic paintings from an 80s magazine of a dancer in very bright, saturated rainbow hues, and about glitch aesthetics. I also thought about a kid with crayons. I painted the central shock of pink, fading to other shades of pink/purple and then a deep blue over the darkest areas. The “synth” got swoops of greens and blues. Very few colors, very little detail. I covered up all the bits that “looked like AI” to me, leaving only vague shapes. I blurred the crap out of it, applied several different procedural (non-AI) processes and filters, did some layering and just kept playing with it — improvising — until I was satisfied. This all was very rapid and I was not at all perfectionist about any of it. Over the top went the text layers I put on the original album art. Done.

This isn’t fine art and I don’t consider myself a visual artist. It’s not trying to say anything with the results, though I suppose I am with the process. The main purpose of the “product” is to give the album a visual identity — but I do think it’s pretty diverting to look at. 🙂

Those who have the album should be able to either redownload the whole album to update the art, or find it at https://f4.bcbits.com/img/a1488757655_10.jpg to manually replace it. Or keep the old version if you don’t care.

I’m not necessarily going to use the same kind of process/thinking on the other replacement covers, but we’ll see. The ones where I used AI tools:

11 Spells: screenshot of a 3D mesh in the software I work on, but then run through an AI-based style filter and then further processed. It’s much more “AI assisted” than AI generated. Also, this album is old enough I don’t really recommend it, so I think I’m just going to let this one be.

No Place: done with Deep Dream or one of the other psychedelic, “everything is made of giraffes and dog’s eyes” early LLMs with a very limited training set. There was a surprising amount of manual work put into it, but I still want to redo it.

Skulk: The figure was generated with one AI tool, then processed several times over by a “restyle” sort of tool, and layered to create the smokey layers. I thought the effect of that was pretty cool. Then I processed a photograph with both AI and non-AI stuff. The problem with the result is, those layers just don’t fit each other very well. I haven’t decided what to do here — maybe I’ll hire someone to paint a cover instead.

Sinister Topography: this was three or four layers of generated imagery, layered together with some manual color painting. I kind of like the look, but I don’t like the idea that the “pencil drawing” stuff should have been done by a human artist. I’ll probably replace the AI with a real topographical map and then see where it goes.

Daydream Network: this was also AI-generated, but I am not unhappy with it. I may replace it just out of principle but it’s the lowest priority.

Infinite Loop: this was AI-generated, and then hand-edited. I like its abstractness, but I feel called to execute something similar by hand-painting it.

Closing Our Eyes…: there are at least three layers of genAI here. I like the unereality of the iris-in-iris-in-iris and the sort of betta fish fins swirling around it. But it’s AI slop and I want to replace it.

Leave a Reply Cancel reply

Discover more from STARTHIEF