I2V-WAN2.2-BlinkBang

<h3>Concept: </h3>

Take a person and move them into a gangbang, being double penetrated.

<h3>Base Prompts:</h3>

On Back

A video of a woman, suddenly cuts to the same woman on her back in a gangbang. Both her anus and vagina are being penetrated at the same time. She is surrounded by naked men. The woman is naked and her nipples, vulva and anus are visible. A man penetrates her vaginally while standing in front of her. Another man penetrates her anus while laying beneath her.

On Front

A video of a woman, suddenly cuts to the same woman on her front in a gangbang. Both her anus and vagina are being penetrated at the same time. She is surrounded by naked men. The woman is naked and her buttocks, vulva and anus are visible. A man penetrates her anus while standing in front of her. Another man penetrates her vaginally while laying beneath her.

<h3>Quirks:</h3>

Finding gangbang videos in portrait is really hard, so the dataset used is entirely landscape. The results in landscape are a lot better than those in portrait and sometimes portrait just outright gives you garbage. The dataset does have a fair chunk of diversity in though so you can do things like specify the skin color of the men to quite a detailed degree.

Also the consistency of the face from before/after isn't always great, partly because in almost any source video the faces are a mess by this point. To help this i'd suggest you use QwenVL on your input image to get a description of the person added to the end of the prompt. I've posted a snippet of a workflow that does this for comfy in the files.

You also don't have to use men, aliens, demons etc all seem to work just fine. Just be descriptive and trust that WAN has some knowledge of it.