Guest
If you like what I do, please consider supporting me on Patreon: We trained StyleGAN3 on a dataset of ~17,000 examples of curated visionary artwork in an attempt to teach a computer to generate psychedelic hallucinations. This was the result. I had tried this experiment previously using styleGAN2 and a smaller dataset of only around ~2000 images. You can see the video for that project here: Interestingly, that particular version of this experiment yielded much more interesting DMT entity-like faces but with formless terrible geometry. In comparison, this version of the experiment did not yield convincing-looking psychedelic entities but does have much more detailed, complex, and structured psychedelic geometry. To be honest, I'm not entirely sure where to take this experiment next. I'm unfamiliar with the newer image generations models that are coming out and the field is moving incredibly fast. If anybody who is well
