This Wan2.2 LoRa makes characters bounce with plenty of jiggles coming from the backs side ;)
Version 1.0 works best when generating bouncing front and back views.
Next version will include more angles with some nice extras.
Trained for Wan2.2 14B t2v high-noise model but also works with the image-to-video high-noise model.
This works with the standard ComfyUI wan2.2 workflow so just connect this lora to the high-noise model.
Note: The trigger word is @ssCl@p but also works without it. Just describe the scene and actions of the characters in natural language
Prompt Template: May help with getting consistent results
Front View:
Front view of a [character] with [feature descriptions]. [Pronoun] does the @ssCl@p while facing the camera. [Pronoun] is in/at [scene description]
Back View:
A [gender] with [hair/face description] is seen from the back. [Pronoun] is wearing [clothing description]. [Gender] does the @ssCl@p with [pronoun] back to the camera. [scene description]
Example prompts:
A woman with black hair is wearing a red latex devil outfit inside a dungeon with torture equipment in the background. She is doing the @ssCl@p while facing the camera.
Closeup of a blonde woman's butt. She leaning forward with her back to the camera. She is wearing tight jean shorts and a small white crop top. She is in a empty bedroom.
Donations: If you want to support my work, you can tip here: https://ko-fi.com/oscai
Description
This Wan LoRa makes characters bounce ;)
Version 1.0 works best when generating bouncing front and back views.
FAQ
Comments (10)
typically you need both high and low noise models to run
Yes you need both models. This lora connects to the high-noise model in a workflow.
oscAI So what do you use for a low-noise model if you need both?
GeorgeCostanza Are you asking if you need a LoRA for the low-noise model of wan2.2? The answer is no.
You do need both main models for wan2.2 for inference. The high noise and low noise one.
@oscAI Right but there's only one model to download here is the point,
@Gradasho Oh, then you only need the high-noise version of this LoRA since it was trained on the high-noise model. Your low-noise wan2.2 model will add the details of your prompt and it doesn't directly depend on the LoRA.
Is this something with your vids or CivitaI? I can see only pictures, wget downloads pictures too or just 0 bytes.
I think Civitai is still processing the videos. I uploaded 4 and only 1 actually plays as a video
it's appears they are "webp" files, which are finicky and stupid and civitai probably only took the first frame. Here's a windows script to turn all webp in a directory into mp4s:
for %%i in (*.webp) do (
mkdir dir%%i
magick %%i -coalesce dir%%i\frames.png
ffmpeg -r 32 -i dir%%i\frames-%%0d.png -c:v libx264 -pix_fmt yuv420p %%i.out.mp4
rmdir /Q /S dir%%i
)
oscAI Some videos get stuck in this weird format forever. Many videos do. I think it has to do with the base format, since it happens a lot to certain people, but not at all to others.