Bursting Out - Wan

The Bursting Out LoRA will teach Wan 2.2 and 2.1 Text-to-Video how to make movies about busty women who have outgrown their clothes. Their breasts are too big for their bras, their sports bras, and their shirts.

Important: This model is 100% fictitious and is not based on images of any real person. All source images were generated from still image checkpoints using Stable Diffusion. No images (neither actual nor AI-generated) of any real person was ever involved in the making of this model.

UPDATE: For the best results, please give my workflow a try. This workflow is what I use to make all my videos using my models and it produces great results in Wan 2.2 using LoRAs that were created with the Wan 2.1 model.

Version 2

Version 2 is more extreme in every way . You can feel free to apply this LoRA to any female character you can think of and it should work. The standard strength of 1.0 will give much stronger results in Wan 2.1 compared with Wan 2.2. For those of you who want unrealistically big, I have a solution for you so keep reading.

The model was trained on images of women wearing bras, sports bras, bikinis, one-piece swimsuits, tank tops, buttoned shirts/blouses, and buttoned sweaters. Scenes with clothing like that will work best with this model.

The training images were extremely exaggerated, making very unrealistically big scenes possible with this LoRA. You can also use it to make much more normal scenes with slightly-too-small clothes and that will work very well, too. Just play with the strength value until you get what you want.

A word of caution: Wan seemed to get a little confused with some of the training data and will occasionally give you a video of a very overweight woman instead of what you might have expected. If that's what you like, then you're in luck! If that kind of result is not what you wanted, your best option is to just use a different seed, although you can try and prompt your way out of it.

Problem: How to boost the strength in the Wan 2.2 workflow to get bigger and bigger busts without ruining the nice Wan 2.2 motion and overburning the details?

Solution: I've cracked the code and will be writing a short article on this, possibly with a simple workflow, as soon as I get time. The answer is to use a 3-part rendering workflow instead of just two, and to use the LoRA strengths in unexpected ways. I'm still experimenting to find out the best combinations, these samples make use of what I've found so far. For a messy "it works but it's ugly" workflow showing this process, use any of my sample videos that have a 3-part strength listing and pull the workflow from that.

The very short summary is that the first part of the workflow uses a plain High Noise model with no Loras at all - not even a Lightx2v or causvid LoRA. That output goes to part two of the workflow which is the normal High Noise section, but here you are free to use speed enhancers and also bump up the LoRA strength as high as you want to get extreme results. Trust me, it will still look good in the end. Finally, that output goes to the usual Low Noise render step and here you will either drop the LoRA strength down to a very low number (like 0.50 or 0.25) or if you're lucky you can remove the LoRA completely at this stage. The result is a video that uses the most possible natural Wan 2.2 motion from part 1, lets you tinker with the LoRA strength all you want in part 2, then part 3 cleans up the mess and makes the output look really nice. You can see the results in the sample videos above.

Prompting:

In addition to the keyword "brstingout", your results can be improved using a few other phrases that were used in training.

• undersized [ clothing item ] - for example, "undersized bra", "undersized bikini", "undersized blouse", "undersized t-shirt", etc.

• Sample sentences:

  • She is wearing an undersized red bikini.

  • She is wearing an undersized cream-colored button-down blouse.

  • She is wearing an undersized white button-down shirt.

  • She is wearing an undersized long blue t-shirt. The t-shirt is tucked into her jeans.

One of my goals in making this LoRA was to be able to use it in my Wan+Vace workflows, as described in my article on Civitai. Whatever your visions are for this model, I'd appreciate it if you post your clips here in this model page or else link this model in your video post so I can see what other people are making with the model. Enjoy!