A LoRA that generates a variety of poses for a solo girl and a dildo.

KREA 2 VERSION

Trained locally on my RTX5090 using ai-toolkit from images of real women with real toys in three separate layers, with number of repeats for each set adjusted to provide balance, as follows:

1. 170 images of "classic" pose of girl on back and legs spread with toy in extreme close-up, close-up and full body shot mixture

2. 48 images of girl on side with knees lifted up together, toy going in from behind

3. 147 images of girl riding dildo under her.

Usage notes:

• No keyword required

• Strength between 0.8 and 1.2 works best.

• Use natural language query based on Qwen guidelines to get the girl posing correctly. Use the prompts in the showcase examples that I have provided if you're struggling.

PONY VERSION

Trained locally on my RTX5090 using kohya_ss from 122 images of real women holding dildo in or near their vagina. Their faces have been whited out to prevent face drift on any character LoRAs you may be using

Notes for use:

• Keyword: dildo_close_up

• Strength: 0.3-0.5. (Going above 0.5 will introduce artefacts from the facial white-out process in training)

The LoRA provides a decent variation in camera and body positioning using these alternative prompts:

1: laying on side, resting on elbow, full body shot, side view,

2: close-up, low angle, laying on back

3. laying on side, ass, showing ass, (my favourite!)

4. on all fours, side rear view, arm leaning back holding dildo (more difficult to realise)

The LoRA tends to randomly have the dildo either in, near or away from the vagina, but you can try influencing using the following prompts:

1: object insertion, vaginal object insertion, holding dildo,

2: imminent insertion, holding dildo,