Is that for the 128k+ with screen swapping? Definitely doesn't look like I'd expect from 48k Spectrums with the chunky attributes but I've not kept up with the shenanigans people have managed to coerce the machine into...
It would be nice to train a neural network to adapt images in a similar way to how ZX artists did it - fitting objects to the boundaries of characters, avoiding attribute clashing and getting pure colors.
