This is a little different, and a little old. I started doing Training because I as looking to make consistent characters for my art - I do stories over at Deviantart.
This is well before I knew about BelieveDiffusion's methodology, which is great, and taught me a lot about quality control... but this is slightly different.
I basically hit upon a random face I liked in a single random Stable Diffusion image, and wondered if there was a way to use it again. I took a standard set of images (I think my Lindi Nunziato's dataset) and used FaceApp to transfer the face from the single image to all 18. I changed the expression a little.
FaceApp is far from perfect, but it did create a selection of consistent faces, which I then trained this embedding on.
That it's based on Lindi Nunziato's body, this means she's got a certain dramatic look, but if you're looking for a Latina bombshell character, here's Valentina.