Another side of my previous lora: https://civarchive.com/models/1587277?modelVersionId=1796171

For generations I’m using 15–30 steps, a cfg of 7, and an 8-12 step shift for videos ranging from 32 to 48+ frames.

The sweet spot for weight seems to be working 0.5-1.0, got some decent results with 0.65. 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.

As usual I just picked some generated images randomly to animate from similar txt2img loras, 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 are some training prompts—feel free to alter it heavily:

A woman with fair skin and short hair, wearing red glasses and red thigh-high stockings, lies on her back on a bed with floral wallpaper in the background. She is revealing her pussy. She uses her hands to spread her vagina, exposing it fully. Her legs are raised, and she looks directly at the camera with parted lips. She has white-polished nails.

A woman with large breasts lies on a black chair outdoors, camera moves down as she uses her hands to spread her pussy, revealing her wet vagina. She continues to spread her pussy with her fingers, exposing her asshole. Close-up view focuses on her pussy.

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.