Trained on:
-3800+ high quality Images from (Avatar: The Way of Water)
-About 15minutes of selected clips from the movie, (focused on characters expressions and body movements)
-500+ Upscaled images FLUX SUPER Workflow , both full images and extracted faces, for diversified character data augmentation
-Images and Videos from my SECRET SAUCE WAN 2.1 dataset, for body augmentation with a color shift applied to limit the human skin color to take over. (This addition to the training was not initially planned, so I may also release the model without these "NSFW" features).
The videos previews are rendered with Lightx2v and Pusa1 LoRAs for faster inference, no other models was used. Using various schedulers / steps / resolution, (Kijai Wrapper)
Using TopazVideo for frame interpolation from 24 to 60fps
You can use these trigger words to force the style:
AVTR Blue skin Neytiri na'vi avatar
I joined a print of the words order as a list with count
Description
FAQ
Comments (6)
it does not work as well in Wan 2.2, will you make a new version for Wan 2.2?
This post contains 2.2 tests, works pretty fine to me https://civitai.com/images/91108969
work very well also with wan2.2
Hi, for 2.2, use this for both high and low noise?
How do I get the characters to appear? Like you have two characters in your previews how do I get them to appear? Is it because I'm using wan 2.2?
It's been a long time since I've used such a well-made Lora! Thank you.
I just a little bit regret the NSFW side, I'm eagerly awaiting the censored version :)