That's a very neat project. The only issue I have with it is that it's basically a passive energy waster. It produces images by burning GPU power, when it could instead curate art from an existing amount of art (of which there is more to ever go through, almost in any category). Some projects that use AI could be replaced with other tech and be much more efficient.
The cost (and environmental impact) of generated images is a rounding error compared to the $7,500+ project cost. As an aside, I wonder how much smaller and faster a diffusion model could be if trained (quantized?) to 4-bit grayscale images.
My home computer batch processing prompts thru Stable Diffusion can generate and then nicely upscale images at at rate of about 1 every 5 seconds. Or 360 per half-hour. Which means a newly image on the display every day to look at for a year in just an half hour of computation.
At about 300W of GPU + 120W of PC, that is 420W * 0.5hrs = 0.210kWhr. This is a rounding error on my monthly electric bill. About six cents.
I've spent more energy than that likely just sitting and reading HackerNews this week.
It doesn't take a lot of GPU to produce one image, and you could always just keep a single image on the wall for a longer time if you want to reduce that impact.
You can also have it not produce images at night or when you are not around the house.
Lots of ways to save energy. The impact of an image every few hours or whatever is nanoscule compared to your transportation and heating needs.
(Also if your apartment uses electric resistive heating, fire away with your GPU, you're just producing images in the process of producing heat instead of passing it through a resistor. It's no less efficient.)
