This is a model used for fixing the ponies drawn by www.thisponydoesnotexist.net in img to img mode. For text to img mode the outcome is not very satisfactory so nvm the text to img mode.
For the img to img mode, use the deepbooru for the prompts and create some negative prompts, set the CFG at 7, denoise at 0.75, and just drop in the original picture generated by thisponydoesnotexist is enough.