This is an entirely AI-generated model.
I started with a handful of poses, a handful of camera angles, and 3 different version of a blank background, which gave me 60 images at 512x512.
I ran it through a little python script which selected the best 32 images.
I used those images to make an Embedding model in Kohya_ss.
Then I ran that Embedding, with the exact same character descriptions as part of the prompt, against a series of pose prompts and came up with about 400 images of the model.
I used the BLIP captioning in Kohya_ss to add character anchors and image descriptions.
I then built the LoRA model with Kohya_ss.
The sample images were all created using an outfits wildcard and the following prompt:
(score_9, score_8_up, score_7_up)
(<lora:SashaCocksl33ve_LoRA:1.0> SashaCocksl33ve_LoRA, SashaCocksl33ve_TI, Short chubby Cuban teen girl, 5 foot 1 inch tall, 38DD breasts, 36 inch waist, 40 inch ass, warm medium tan skin with golden undertones, even complexion, high melanin skin, oval face, brown eyes, thick eyebrows, button nose, full lips with fuller upper lip, very long curly hair extending all the way down her back, soft curvy body, realistic proportions, slightly chubby body, soft feminine build, natural curves, realistic anatomy, no exaggeration)
(__availability_outfit_styles__)
Negative prompt:
(score_6, score_5, score_4, score_3, score_2, score_1)
Images are not cherrypicked.
A total of 40 images were created with this prompt.
The first 20 were added to the Embedding model's page and the second 40 images were added to the LoRA model's page.
I would suggest adding prompts or negative prompts for hands, which I did not do here.
I used very limited prompts in this example to illustrate the stability of the model.
The Embedding - SashaCocksl33ve_TI
https://civarchive.com/models/2242390?modelVersionId=2524958
The LoRA - SashaCocksl33ve_LoRA
https://civarchive.com/models/2242947?modelVersionId=2524971
While this character is entirely AI-generated, it is based off of an online friend of mine that was too shy to send me her pictures to make a model with.
Instead, she gave me the detailed description of herself (except for the word "teen") that is used in the prompt above, though I did add a few details after talking to her further.
Of the hundreds of test prompts that I ran after finishing making the models, she referenced a handful of them that look exactly like her, according to her.