Upload 73c580554d0248122a2d83b281d100d2.safetensors
Browse filesLoRA trained on videos of women taking off their clothes. Pulling shirts over their head, unzipping onesies, flashing trenchcoats, pulling down pants and dresses, it has a bit of everything.
Wan T2V 14B Version
I was pleasantly surprised at how easily Wan follows prompts. You can get what you want from 0.5 strength up to 1.0. However, the base model is pretty crap at nudity and has really strange concepts of genitalia. I recommend using a more general nudity lora at low strength (0.1-0.2) to compensate. I haven't seen one available yet for 14B, but there is a female masturbation one that can do the trick.
I've added another post testing my lora with the I2V 480P model, and it seems to work pretty well. Try it out!
Hunyuan Versions
I tried a very lazy approach with this one, automating as much of the process as I could. As a result the captioning is pretty bad, and it will take some finessing to get what you want.
Add the tag "str1p" at the beginning and end of your prompt, describe the scene in full setences and add booru tags with what you want to see. If no ?????????? occurs in the first attempt, add nudity tags like "breast", "pussy", or "nude". If the video starts naked, describe the clothing a bit more. For inspiration you can drag the posted videos into ComfyUI or check out the .txt captions that were used for training.
Some aspects of the automation I think are worthwhile, so I've posted the workflows.
Version 2 just has the same videos with better captioning, which makes the lora a little easier to work with.
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9deedf78630532df8fc2fdda98020c39354a93ea4f5cfffdfdb92d600e914114
|
| 3 |
+
size 306807976
|