The author of the model is very good at compiling prompts and selecting the parameters of the neural network. But the model itself does not carry anything good. Just to make the model work, you need to break at least 3 tambourines. Custom values for all parameters, including eta and clip skip. And you get an average result. BUT the question is why such perversions, if you can get better results on other models, making less effort.
On standard parameters, nothing special turns out, but with specific ones, you can make beauty out of any model. The model has no flair.
All of the above is solely my personal opinion and is not an unconditional statement.