Taking a long break brings knowledge. So I updated the dataset and finally released a WAN version. If using another version refer to the details for that version. Gallery images have prompts but using words like "They are engaged in tribadism" or "Their legs are interlocked" and "They are rubbing their pussies together" works. Also using different samplers, schedulers, cfg, and speed loras can affect results. I use simple with euler with light2x speed loras.
Description
ALPHA (HUNYUAN)
Trigger words: They are engaged in tribadism, with the two women rubbing their pussies together. scissoring.
Strength: 1.6
Steps: 7 (I've only tested this on fast hunyuan but it should work on all types)
Shift: 6
Guidance: 7 - 9 (I use 8 mostly)
Length: 49 - 109 frames
Dataset consists of 15 videos at 24fps @480 by 288. Does well at landscape resolutions. Hit or miss doing square or tall/portrait resolutions. I've learned a lot with this lora and I will be adjusting my training steps after this. This is why the lora strength needs to be high as of now. Each image should have metadata so you can drag and drop it. Only tested on Fast Hunyuan.
FAQ
Comments (1)
scissor me timberz!