This is an unconditioned diffusion model trained on around 500 miscellaneous tiki images from around the web.
The resolution is 64x64.
It was trained for around 4k epochs with a loss around 1.5%.