v0.2 Notes:

Tweaked training procedure. Increased dataset by 12 times, more angles and variations. Seems following prompts better and more stable in my case. Added some optional buzz begging for better training setups xD

Tagged whole dataset for v0.2, now it is more concise, and little bit more human like. Old prompts should work but I recommend the new style:

A woman with light skin and long blonde hair exposing her ???????????? and ass. She is looking over her shoulder with a slight smile. She moves her hand grip her thighs to spread her ass.

For generations I’m using 15–30 steps, a cfg of 6, and an 8-10 step shift for videos ranging from 32 to ??+ frames. Higher shift seems to be triggering the sitting effect faster in my case.

The sweet spot for weight seems to be working 0.5-1.0, got some decent results with 0.65 but I have to say this one was tricky, sometimes it just creates body horror. Sometimes you just need to be lucky or have good conditioning starter frame. It works better if the initial image fits into to the lora concept, might alter this in future versions...

As usual I just picked some generated images randomly to animate from trending page, kudos to creators.

No trigger word used for this time for experimenting purposes. I recommend focusing on the action rather than the scene since this is I2V (at least that works for me).

Here’s an example prompt I used in some of these generations—feel free to alter it heavily:

A woman standing as her back turned towards camera. Her hands on her thighs as she grabs them for stretching her large, round buttocks making her vulva exposed.

A woman with fair skin and long brown hair lies on a bed with white sheets, wearing a white, partially visible top. She is ????? from the waist down, with her legs spread and raised, exposing her vulva and ????. She has dark red lipstick and black nail polish. Her right hand is on her right buttock, and her left hand is on her left buttock. The video focuses on her genitalia.

etc.

I’m still experimenting the training process, so feel free to share your generations. This will help me gather more feedback to improve further. I didn’t have a large dataset to begin with and trained on relatively low settings if anyone’s curious.

Since this is an NSFW-capable model, please use it responsibly. No real person faces were used in the training.