Disclaimer - This is more like a v1 beta of the model. It could definitely use improvement; However, it's possible to produce good results with the right model and/or prompt.
Weights between 0.85 - 0.95 seem to work best
Best prompt prefix I've found is:
HaiStei, a photo of a woman,
Model trained on the base SDXL 1.0.
Best results I've found have been from:
Model: Dreamshaper XL
Sampler: DPM++ 3M SDE Karras