Sai is a versatile model, trained on only face images (portrait) so it works well on flexyble prompts
A lora trained on images made by IMG2IMG with ForgeUI