Premium character LoRAs
Original synthetic characters, bundles and AI tools are in my shop:
Join the community on Discord: https://discord.gg/nB79wRNtMT
Face Sitting
A concept LoRA for Krea 2 focused on facesitting: a woman straddling or squatting over a partner's face, facing forward or reverse, shot from above or from the side. It gives an amateur photo look.
Usage
Start with the trigger word FaceSittingRK, then describe who sits on whose face, how she sits (straddling, squatting, facing forward or reverse, leaning back or forward), what shows of the person below, what is exposed in plain words, what she wears or pulls aside, the camera angle and the place, and end with the light. Naming the position and the angle brings out the effect the most.
FaceSittingRK, Photograph of a woman with <hair>, <nude / in panties pulled aside>, sitting on a man's face, <facing forward / reverse>, <view from above / POV> on a <bed>. The lighting is <light>.
Recommended weight: 1. Made for Krea 2 Turbo (8 steps, CFG 1).
Prompt Examples
"FaceSittingRK, Photograph of a nude woman with fair skin and dark hair, sitting on a man's face. His face is buried between her thighs, with his mouth near her vulva."
"FaceSittingRK, Photograph of a nude woman with fair skin and brown hair, viewed from behind, straddling a man's face. The man, with short brown hair, is positioned on his back with his face between her legs, his mouth near her vulva."
"FaceSittingRK, Photograph of a woman with fair skin and long black hair with brown highlights, wearing glasses and dark lipstick. She is straddling a man, whose face is obscured as she sits on him, with her hands in her hair, eyes closed, and mouth slightly open."
Training
- Base model: Krea 2
- Type: Concept LoRA
- Steps: 4000
- Rank: 32
- Optimizer: adamw, constant learning rate
- Engine: ai-toolkit
Generation settings
Trained on Krea 2, and works on Krea 2 Turbo for faster generation.
- Raw: ~30 steps, guidance 3.5
- Turbo: ~8 steps, guidance 1.0
The gallery images were generated with krea2_turbo_mxfp8.safetensors, at 8 steps, guidance 1, LoRA weight 1.