WAN
Trained on a small dataset. Use "One woman is sitting behind the other woman, reaches around her hand and is fingering the other woman's pussy from behind" for the trigger. Use a strength of 1 - 1.3, I've noticed with more complex scenes it can struggle so up the strength can help. Struggles with landscape shots, decent at 1:1/square shots, better at portrait shots.
Gallery images has metadata for exact settings and prompts.
For other versions click on the "About this version" for trigger words and descriptions.
Description
V2
Half the lora size, clip skip 2, upscaling some of the lower res images in the dataset, and changing the trigger words from V1.
Add "sitting", "lying", or "kneeling" for different positions.
Excels at tall images and square aspect ratios and struggles a bit on landscape/wide ratios. I find using "Spread legs" on square and wide aspect ratios can help a lot with anatomy. Adding "Full body" works well but sometimes can mess the anatomy.
Will come back for a V3 and add more wide images in the dataset
Trigger words: r3a3hk1ck1x, girl fingering another girl from behind, reach-around, 2girls (2girls is baked in but adding this might help)
Optional words: sitting, lying, kneeling, full body, spread legs
Strength: 1