The face looks better now, the colors are more restrained, and the expression is more enhanced.

This model is the 1.5 version of my SDXL Colorful Anime model. The original intention of creating this model was that many people's computers couldn't handle the SDXL version, but they liked this style a lot. Initially, I didn't plan to create it, but seeing many people hoping for it, I decided to release it. However, the first version I uploaded had poor results, so I released 2.0. Then I realized that I needed special negative word constraints to generate good images. So, I made version 3.0. The previous version required negative word constraints, which resulted in poor usability. This version fixes that issue. The usability has improved, and overfitting is no longer an issue.

This model is entirely designed in the style of SDXL. Some of the Colorful Anime images from SDXL can be recreated, with a background in the colorful crayon style. Compared to the regular 1.5 model, the characters have a much better texture.

Recommendations for image generation: The sampling I used for image generation testing includes DPM++ 2M, Euler a, and DDIM. For DDIM, you need to increase the sampling steps. For the other samplers, I used 20 steps. The size I recognized using is 512*768*2 .540*960*2. For upscaling, I used R-ESRGAN 4x+ Anime6B.