Trained on 20 pictures of mostly naked women with large breasts laying down in bed. This model works best when using img2img. It can be hit or miss when using plain text-to-image.

Technical: Prompts can be anything you like and do not have to include "bed." I suggest starting off with my 512x512 example images as your input to get the ball rolling. 80% of training data was light-skinned woman, 20% dark-skinned, so you will have better quality when generating lighter skinned women.

Bad Generations? Specifying the desired ethnicity is helpful, especially for darker skin tones. If nipples are missing, describe them more in prompt. If areolas are weird shapes, use better input image or add "round areolas" to prompt. If eyes are on nipples, specify "eyes on nipples" or "eyes on breasts" in negative prompt.

It is good at making:

• Women with huge boobs

• Large areolas

• Women with smaller boobs (with proper image input)

It is bad at/untrained for making:

• Perfect fingers

• Multiple women

• Very different angles than top-down

• Whole bodies

Example positive prompts: naked woman laying down, lying down, large breasts, round areolas

Example negative prompts: extra breasts, extra nipples, clothes, glasses (if applicable)