If you're here, you probably like ????????. This model excel at women presenting you their backside.

My motto: Quality over quantity. The low noise model has been trained at the native 1280x720 resolution of the dataset for maximum details. The assembled dataset is very sharp, high quality videos that were handpicked and carefully cut.

The model seems to have learned more about ??????? than the intended concept. Explore and use your creativity. The last few provided samples highlight such case.

Generate and please share your results.

What it can do

• Close-ups, medium shots, distant shots
• Standing, kneeling, laying, doing the split
• Mostly backside views, snuck some frontal views in it
• No public hair / With pubic hair

Dataset

• Number of videos: 143
• Resolution: 1280x720
• Frame rate: 16
• Duration: 5s (81 frames)

Training

High noise

• 10x A40 GPUs
• 845x475 resolution
• 40 epochs
• 0.0001 learning rate

Low noise

• 8x A100 SXM GPUs
• 1280x720 resolution (native dataset resolution)
• 37 epochs
• 0.0001 learning rate