This model was trained on Foxyreine images. The dataset consists of 47 images, 10 steps.

Basically, the main problem is the scarcity of available images, mostly all images are portraits or cowboy shots, because of this, it is necessary to strengthen the distance and composition tags. (Full shot, full body etc.) x 1.2 - 1.4.

You can also enhance the style using the author's tag (@foxyreine), which is included in the ANIMA. BUT, watermarks will appear.

I also primarily used a combination of loras (anima-highres-aesthetic-boost and anima-turbo-lora-v0.1) and better poses from the ANIMA developer to speed up the model. Because of this, the results of standard generation can be poor. Experimentation is required here.

Good luck.