This model creates generations with a hexagonal composition. You must include "hexagonal composition" and "simple background" to get the full effect. Recommended strength is 0.6 but a wide variety of strengths work.

Training data is entirely synthetic, with version 0.1 only using 6 full images. As such, there's a good bias towards elements from the originals when the strength is high.

Uses a technique wherein training is done using both high resolution base images and downscaled versions at half resolution. This works exceptionally well when you start at the native low resolution (512x512 or ??0x??0) and upscale by 2x (using img2img or high res fix).

Trained on Galena v3.0, but should work fine on many other models.