Dataset creation:
Trained on 128 HiRes (at least 1152x1536) images created local SD a1111 with ZavyChromaXL 1.0/2.1 model.
Dataset overview + samples:
The model works with all kind of prompts and even simple prompts very well. With simple prompts, persons can have the tendency to "long neck", "stretched torso, arms, legs". To put these in negative prompt then is recommended. Works good with other models. Transfer your img to this style: reduce your prompt to only subj/background related. Then evolve with the LoRA applied from 0% noise upwards. Great results.
Comments