This community is for people who train their own image generating models, to share their output images and videos, but until more people volunteer to do that I'll continue to post my own. I hope that's okay, and thank you for your interest!

So I tried an experiment recently, where I tried to reproduce my artwork simply by using stable diffusion prompts, rather than with starter images, or my own trained models. The results, though not truly representational, are pretty interesting!

This image was rendered from a GAN model trained from 1000's of images of my own natural process painting work. Since the process renders organic abstractions with fractal imagery, there is a large amount of inherent zoom possible for obtaining details without losing continuity, and that zoom is only really limited by the dimensions of the work surface, and the grain of subsequent pigment layers (too much zoom and the color begins to look like a grainy photograph).

Because of this, I can get up to 30-40 images or more from just one painting by simply varying the level of zoom. This is very important when you need 2-5 thousand images to train from!

Made with Nvidia Playground.

This image was produced by a GAN model which was trained on thousands of images of my own artwork. Because this work is fractal-based, and uses sediment deposition to apply pigment, the level of zoom that can be achieved is much higher than with conventional artwork, allowing a person to take multiple detail images of one single painting.