Concept

A woman leading a man by his penis.

Sometimes, you see a video and it just hits that CFNM/Femdom goon section of your brain right. Well that happened to me with a video of a woman pulling a dude along with by cock. Unfortunately I quickly found very very few videos of this exist. I eventually found a whole 26 videos and trained this Lora. Unfortunately given the limited number of videos the POV concepts are quite hit and miss, and might require some seed hunting. If anyone has any suggestions where I might find more source material I'm happy to add it and re-train this.

Usage

Should work in both T2V and I2V. Base prompt is:

A video of a woman leading a man by his penis. The woman grips the man's penis with her hand and pulls him along by it. She walks to the <right/left> of the scene with the man following behind her.

Technically the model is trained with female and male POV videos, but the dataset for those is tiny so I'll provide the prompts but do not suggest it'll work well.

Female POV:

A video of a woman leading a man by his penis. The video is from the woman's point of view. The woman's arm appears from the right side of the scene. The woman grips the man's penis with her hand. Throughout the video, the woman's hand remains on the man's penis. only the woman's arm is visible.The woman walks towards the viewer and the man follows her.

Male POV (I would only suggest you do this with I2V where the man's body is already in the frame. T2V will be a horror show.):

A video of a woman leading a man by his penis. The video is from the man's point of view. The woman's arm appears from the top of the scene. The woman grips the man's penis with her hand. Throughout the video, the woman's hand remains gripping the man's penis. The woman walks away from the viewer and the man follows her.The camera moves with the man. The man's hands remain at his sides. In the final frame we can see the woman's hand gripping the man's penis and the back of the woman's body.

For these test videos I've only used this Lora but for better results using something that improves penis shape is probably a good idea.

Training

The model was trained on 26 videos, 18 non-pov, 5 female pov and 3 male pov. all 5 seconds long and 16 fps. High noise was trained at 320 resolution with 4 frame buckets. Low noise was 704 resolution with 2 frame buckets. I included 4 images because I happened to have them, they were trained at 0.6mp and 1mp respectively for high/low. 50 Epochs at 4 batch for High, 50 epochs and 2 batch for low. All trained on diffusion pipe on 4x3090s. Captioning was automated using JoyCaption and DarkAges Llama 70B.