I’ll start with a teaser for next week, though. This picture on the right is one she gave me in response to the prompt “Kitchen Table Kibitzing painted by van Gogh”. I chose it because it best matches, in color and style, my little van Gogh weekly newspaper opener above the fold. You can see that craiyon is experiencing a plot about van Gogh’s style and color palette, but he doesn’t know, for example, how many legs the furniture would likely have. I’m kicking off next week’s thinking about why that would be.
What I have this week is what I found while searching for next week’s explainer videos: a group of YouTubers who have made a habit of illustrating popular song lyrics with art generated by AI. Most seem to use Mid Road, which was publicly available before craiyon. Midjourney requires an account, which may or may not be available at any given time due to the volume of requests.
Most of these videos seem to share a look, some sort of fantasy/scifi/metal tone, most of the time. I don’t know how inherent this is to Midjourney – I guess it’s not much, but that’s just a guess. It wouldn’t surprise me if it’s just that people who want to do this come with a similar mindset.
All manufacturers declare that the prompts given to the AI are the lyrics of the songs, but they cannot be just song lyrics, frame-by-frame, or visual style wouldn’t be as consistent, even in images that aren’t have any words. Some kind of extra incentive is given, although we understand that the AI prefers the particular style we see.
I’m going to give you an example. This video, Don’t fear the reaper by Blue Oyster Cult, is Daara, which specifically credits Midjourney for their footage. Take a look at the images that appear at the start of each verse, when the guitar vamp takes over the percussion. You can’t parse the one at the start, but for the next two verses the AI helpfully provides some puns, I’m sure it’s not intentional. For good measure, he decorates the third with a few doodles he thinks are related: “COWBE”. There’s no way the machine knows, by itself, from words alone, that “cowbell” has something to do with this song. [5:10]
OK, I’ll shut up now and let you watch, or not. TheloSpike brings us Queen’s Bohemian Rhapsody. [6:05]
Of don’t help: David Bowie Spatial oddity. [4:51]
From Daara again: The Eagles Hotel California. (Everything Daara does seems to particularly encourage the AI to add its mutant lettering to images.) [6:25]
Of Slow RF: Note that this is NOT Metallica’s Enter Sandman – you can see this song here on Daara’s channel. No, it’s the snitch 1954 Mr Sandmanby my Sweet Adeline sisters, The Chordettes. This song fights against the preferred, rather pompous visual style, sometimes to hilarious effect (see: “Make him the cutest that I’ve ever saw”). [2:20]
In jetboyFrank Sinatra’s contribution take me to the moonwe can see that the manufacturer did not give the AI the same impression of apprehension as Mr Sandman apparently was assigned. [2:29]
TTV sawfish: One would expect the more typical style that we have seen from the beginning for Pink Floyd Time, not a happy song. This manufacturer seems to have taken a different overall approach, however, and kept the album art front and center when inciting images. [6:56]
SolarProphet:ELO Mr. Blue Sky is also visually more cheerful, but how quickly we get back to type when M. Night shows up! [3:44]
Kujo: King Crimson The Court of the Crimson King opens with album cover (someone on the internet published an AI-generated copy. It was… not a faithful reproduction), but quickly moves into AI Lord of the Rings book cover territory. [2:09]
Of LoFi Farquaad: At Simon & Garfunkel The sound of silence. [3:09]
De Daara: Always the final, that of Led Zeppelin stairway to Heaven. [8:05]