site banner

Friday Fun Thread for November 4, 2022

Be advised; this thread is not for serious in depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

4
Jump in the discussion.

No email address required.

Does anyone here fiddle with AI art in their spare time? I need to make some fliers in a concert I'm playing, and I want to generate some AI art for them. I'm very limited on time this next week and don't have time to learn and generate images for it so I'm looking to outsource some of the work.

I've dabbled with it a fair amount, and am not a furry lol

What sort of stuff you thinking of? Stable Diffusion is a lot better at some subject matter than others.

I'm thinking of a cubist styled picture of a clarinet, oboe, and bassoon. If people are playing the instruments, they should be pretty undefined. It fits with the style of music we're playing along with the time period a lot of the pieces we're playing were composed.

Interesting challenge, this! The difficulty with SD right now is coherency; it's best at humans and landscapes bc these are the most common subjects in the training data. Even then it often gives people too many limbs, fingers, joints in the wrong spot, etc. So something as intricate as a locomotive, a gun, or a woodwind instrument it has little prayer of being rendered with full accuracy. It can make tubular things with lots of valves and circles and the occasional trumpeted end, but not anything a musician would mistake for a real-world instrument.

So I've had to try going pretty abstract, getting it to *suggest *the instruments in cubism chaos. Even then it's not easily applying cubist logic to the instruments themselves -- instead simply depicting them realistically atop a cubist background, or having the instruments simply devolve into abstraction. Quite interesting! This was the struggle the cubist masters themselves struggled with - the fight between abstraction and meaning!

Anyways I have a few ideas about how I might get something better with prompt modelling - where you have the AI change what it's trying to generate partway through making an image - but for now here's some gens I made. Most are pretty rough, but I erred on the side of inclusion as I don't understand cubism very well. If you like something and need it higher-res or something, let me know.

https://imgur.com/a/P1gPPdu

I really appreciate it! I think a few of these will work well - This is a small enough concert with a short enough timeframe that I'm not too concerned about quality or artistic accuracy. I like the surrealist nature of a lot of the generations thought I can see your point that it seems a lot of images have the instruments interposed over the cubism instead of being integrated into the image. I'm not an art (or at least this type of art) aficionado myself so my understanding of cubism is probably as pedestrian as yours. I'll message you for higher res in a bit.

All good on this? Haven't seen any messages but I've never gotten any on this site so I could be missing it, so just making sure lol

Depending on the handout could have easily been enough resolution to start with

Yes the innate resolution was fine, im just putting on standard paper.

Thank you so much, some of those are exactly what I was looking for!

The easiest way to get the results you want is to get a photograph of your musicians and run it through a free online style transfer app with keywords like cubism and picasso.