---
license: apache-2.0
base_model:
- Wan-AI/Wan2.2-T2V-A14B
tags:
- not-for-all-audiences
- wan-2.2
- lora
- diffusers
- template:sd-lora
pipeline_tag: text-to-video
library_name: diffusers
---

# Wan2.2 Scat Lora

## Info

UPDATE: 03-25-2026 New Version: ScatMaster-V2 Released

```bash/powershell
pip install huggingface_hub
hf download obsxrver/wan2.2-scat-xviii-PooMaestro --local-dir ./wan2.2-scat-xviii-PooMaestro
```

### Activation Words (non-exhaustive)
> she defecates

> covered in feces

### Sample Prompts
> A woman, she is nude, lying on her back, looking at the viewer, legs up, covered in feces, she is rubbing her pussy
 
> nude woman, squatting over a bed, she defecates on the bed, looking at the viewer, moaning


Example videos were generated using Epoch 75 for high noise and Epoch 85 for low noise. 
An example workflow available to download. 

## Q&A
1. > Q: Why did you do this?
   
   > A: Because I can, next.
2. > Q: Dataset?

   > A: My dataset consisted of 33 ***high-quality*** videos and 16 images, some of which were stills from the videos. For the videos, I trained with a size of 298px^2 (0.0888MP), and the pictures at 960^2 (0.92MP). The dataset needs to be high quality and free of noise or artifacts. Otherwise, these will translate into artifacts in the final LoRA, which is no bueno.
    
3. > Q: Captions?
   
   > A: I captioned these using Gemini 2.5 Pro, each caption consisted of "[Setting], [Subject], [Action], [Lighting]" The exact system prompt is available in the repo. I manually added "still-frame" to every image(not sure if this helps), and I also blurred out the actor's faces in many of the videos, and added "her face is blurred out", to the caption. This helps with negating the undesired effect of baking-in the shape and visual characteristics of any given actor's face from the dataset into the LoRA's output.
   >
   > Example: A room, a woman wearing a grey shirt, her face is blurred out, squatting on the floor and spreading her ass, a plain wall, the woman defecates, bright, direct lighting.
4. > Q: Training?
   
   > A: I used Musubi-Tuner. You can find the training commands outlined at [https://github.com/obsxrver/wan22-lora-training](https://github.com/obsxrver/wan22-lora-training). I use these training commands for every LoRA I make and almost always have good results. 
#### Be smart, use common sense.

# I am not responsible for your actions