v1.5: trigger word changed to <code>t4k3sh1m4_3ku_style</code>make the model slightly less wonky with some type of images, but it is still better to post processing those images for better qualityv1.0trigger word: <code>takeshima_eku_style</code>known problem:<ul><li><s>when creating full_body character (head to toes, standing, direct view), you would get really bad face, this was due to biased training data, please manually fix by using hi-res fix, or cutout and upscale to img2img, or any other method (refer to the last example) </s>turned out the original model also isn't really good with this particular type of image, so a lora wouldn't be able to change that</li><li>big_breasts tag destroyed the shading style, for whatever reason lol</li></ul>

Files with SHA256: dd0ac0b3066309e531a663f096880606c10bcf2d3ad9dfddbed055927a383fbf