Ensure your class word is in all of the prompts. You can use my 'Smart Process' extension to do a rough job of captioning images, then go through and fix things if they're way off base. I think they help a lot, as well as training with text encoder. Appx 5-10 each should be enough, but again, more is better, IMHO.Ĭaptions.
Go for a mix of face closeups, torso+face, and full body for best results. Make sure they aren't blurry, try to avoid cropping parts of the face, stuff like that. IDK about a 'good' guide, but you can probably try something like so:
