It may not seem like it, but you have a huge following. Don't stop making these models.

Base SD XL checkpoints are rare so I thought I would try this one out. The opening crowd image really impressed me. On the positive side, it has very good face variation and seems to handle fingers and toes well. But you absolutely must follow the author’s specific steps and CFG. If you normally use these values to tweak output – well its not happening. I would also suggest using the normal scheduler rather than the Karras. I tested multiple prompts and resolutions because I do like many of its generations, but it has other issues such as hyper “crunchy” details that ruin the output. For example: “a motorcycle racing down the Los Angeles coast” adds hundreds of cargo containers embedded into the hillside, wood has an artificial appearance, cloth can look hedge maze and skin can have too much detail. Add in the over sized eyes (starting at CFG 3) and you have a variation of the uncanny valley that is unique to AI. If you are looking for a better general purpose SD XL model, I would recommend the Epic Realism, Juggernaut or Stable Yogi models. Note, my review is based on output from ComfyUI which could be enhancing these issues.