



















Trained on 200 images of pixel art scenery. I've only tested it on the AnyLoRA checkpoint and i would recommend using it. Use a weight around 0.7 and a CFG Scale of 7 for best results.
I usually generate an image at 768x512 (or 768x432 for a 16:9 image) and use hires fix at 0.5 denoise, 10 steps and with the R-ESRGAN 4x+ Anime6B upscaler. In addition to that I use the PXL8 extension to downscale the image and reduce the color palette. If I am generating a 768x512 image with hires fix i usually downscale it by 4x to end up with a 384x256 image, I usually put the color palette size at 64. Right now it struggles with character, sometimes it works if they are further away.
I would recommend using "scenery, fantasy" mainly. Otherwise just look around on danbooru or other booru sites for tags, some nice ones are "dappled sunlight", "mountainous horizon", "overgrown" etc.
1. 转载模型仅供学习与交流分享,其版权及最终解释权归原作者。
2. 模型原作者如需认领模型,请通过官方渠道联系海艺AI工作人员进行认证。我们致力于保护每一位创作者的权益。 点击去认领
