toyxyzさんのイラストまとめ

@atroyn For this tool to function properly, the images used for training must be separated and presented by subject, like B rather than A.

@atroyn Just searching for and showing visually similar images can give people a false perception.

https://t.co/HTUmPcXsZo

Outpainting test #stableDifusion #AIイラスト #AIart

U-Net Blocks Weight Merge test #stablediffusion #AIイラスト #AIart #txt2img

U-Net Blocks Weight Merge test #stablediffusion #AIイラスト #AIart #img2img

Depth model Img2Img test #stablediffusion #depth #AIイラスト #img2img

Since details are randomly generated with only a flat shape, it is most appropriate to use a 3D model to specify a precise pose. #depth #stablediffusion #AIイラスト #img2img

Using the Danbooru tag for WD and the textual prompt of the general SD model at the same time can be helpful in obtaining the desired image. Original / Above: no SD prompt / Below: with SD prompt

If the ratio of either WD or Depth is over 1.0, the image is completely destroyed.