A video of a woman, suddenly cuts to the same woman naked. The woman is embarrassed by her nudity and is trying to cover herself with her hands.
It seems to struggle with certain types of clothing like bikinis or underwear, so you might need to add:
Her nipples, vulva and vagina are visible.
If you want the person to be looking at you during the video add this:
She looks at the viewer for the duration of the video.
Many of the helper loras for genitals or bodies work well this as the motion was more the focus of the training. Also it seems to love going bigger up top so you might need a lora to counteract that.
Weight of 1.0 for both high and low should work well. Sampler wise most should work but i'd avoid unipc as it does some weird things. Stick with Euler/LCM/Res.
Training
Trained at 0.1 MP @ 4 frames for high and 0.35 MP @ 2 frames for low. Batch of 4 on high and 2 on low. Dataset used was 47 videos. Captioning was automated using JoyCaption and DarkAges Llama 70B. Trained on diffusion pipe on 3x3090s and 1x5060ti.
