WIP E-Girl style Wan2.2 low noise lora
Intended to be used with Phr00t's AIO T2V model V10 https://huggingface.co/Phr00t/WAN2.2-14B-Rapid-AllInOne
Description
FAQ
Comments (9)
i bet she cant hit the whip
Low noise only because you use only 1 sampler at workflow? Or you use 2 samplers but this Lora needs only for low to work?
description says intended for phr00t's AIO so that's using a one sampler merge. whether you can call that wan 2.2 is debatable.
@shinizaya That's true. However, the Lora is trained on the base wan 2.2 low noise model so it's not specifically for AIO, it's just what I use and seem to get decent results with
@Jess_ai_ca Cool, I use Rapid AiO. But now there is so many versions of it. I tried v10 and Mega v1 for this Lora, works with each but results are so different. Curious which version you use?
@flo11ok874 I use v10. The new Mega versions seem to give terrible T2V results for me
Interesting that I usually get better results in AIO (4steps) than normal 2.2 with high noise and low noise (2 steps each). Also, VRAM is used better somehow, with rarer OOMs. The only problem is with Start and End Frame (FLF2V). You need a clipVision loader in each image separatelly, and even so, there are times it wont make a good transition. If you have 81 frames, for example, 79 will be frame1, and 2 with frame2 in 30% of the time.
@Jess_ai_ca If I use stock wan, would that explain why my girls faces don't look like your examples vids? just learning about this...
Is this meant as a character lora or a egirl aesthetic lora to be applied on top of a subject?