Ripped Clothing - Wan

Ripped Clothing for Wan

This LoRA will allow you to add rips, tears, and frayed edges to different parts of people's clothing. Wan knows how to add holes in clothing, in a limited way, but the holes usually don't look natural and Wan likes to clean up the edges of the holes as if the gaps were hemmed in a shop. This LoRA was trained with many images generated in Z-Image-Turbo, which natively does a great job of creating ripped and frayed clothes. While all of the training images were of women (sorry, habit) the LoRA does work with male characters as seen in one of my sample videos.

Important: This model is 100% fictitious and is not based on images or videos of any real person. In fact, every video in my dataset was created in AI with the heads out of frame to avoid biasing the faces. No images or videos (neither actual nor AI-generated) of any real person were ever involved in the making of this model. LoRA dataset available on request.

The Low Noise model is more important here than the High Noise model. In most cases, you can get very good results by using only the Low Noise model at 1.00 strength. This will produce the least amount of change to the motion and overall look of your base video. You will need to experiment on your own with the two models and decide for yourself what look you prefer. All of the sample videos were created with both High and Low noise set to 1.00 unless the video caption says otherwise. Remember that Wan is generally responsive to prompt guidance, so be descriptive about what you want and Wan will usually help. See prompt wording guidance below.

All of the training images had the faces blanked out, so the model SHOULD change very little of your character's faces, although I do see some changes when I directly compare with and without the model. See the comparison video in the samples above. The training images had lots of different hair and hair styles, so you're more likely to see hair changes when you use this model. The training set also included several muscular/athletic women, since that often goes with the ripped clothing theme, so this model will respond well to prompts for women of that body type.

I've tested the model on top of some of my other LoRAs and it seems to combine well, although you may need to change the strength of the Ripped Clothing LoRA when you mix it with other LoRAs in order to get the effect you want. Again, see some of the samples above. I have tried making some videos where the character actively rips her clothing with her hands, and I've had limited success with that. I can get it to work, but it takes several tries and careful prompting. I may have better luck with image-to-video, so I'll give that a try later and post results.

Prompt Guidance

This LoRA does not have a specific trigger word, but I used a few key phrases consistently in my training captions. In my captions, I usually set up my prompt like this:

"A woman is posing in front of a plain background. She is facing the camera with her hands on her hips. Her blouse is ripped and frayed. Her short skirt is ripped and frayed. Her tights are ripped."

If you want to add rips and tears to a specific part of the clothing, you should describe the full outfit and then in a separate sentence, you can add the rips where you want them, as in "her blouse is ripped on the shoulders" or "her tights are ripped on the thighs". See the sample video of the goth girl on the bus for an example where only the tights are prompted for rips.

In my captions, I used the word "frayed" when the image included frayed edges at either the bottom hem, the neckline, or the sleeves. If all the edges were clean, I only described the clothing as "ripped". I don't know how well Wan picked up on that detail, but you can try it if you want in your own videos.

If you describe the whole outfit with the word "ripped" in front, then you basically give Wan free rein to add tears wherever it feels like. For example, if you use an outfit wildcard, you might say "she is wearing a ripped __female_outfits__" and you will get whatever Wan gives you, although you can guide it a little by describing patches of bare skin on certain parts of the body and so on.

I hope you enjoy working with this model, and please remember to flag the model when you post videos so I can see what people are making.