This is a retrain of my model but on Dreamlike Photoreal. (Please see their terms if you want to use this or a model you train off of it commercially.) Anyways, after hearing that F222 was supposedly trained on only 11 images in Dreambooth, I decided to try retraining my model again using the 11 most distinct images of the original dataset. The facial quality wasn't as great so I added another 11 faces that I handpicked.

The prompt I trained this model on is: barefoot woman standing in her panties, but you don't necessarily have to use that to get similar results as the example pictures since it can be a bit awkward to put into some long and creative prompts. Just mention several of the keywords from the phrase and you should still get decent results. If the faces aren't great, add hot face to your prompt, i.e. barefoot woman standing in her panties, hot face which will increase focus on the face (but decrease focus on other parts as a tradeoff.)

(Note: after the recent update, use an initial resolution of 384x768, an upscaling factor of 1.33333, and an upscaler of Latent.)

0 comment