[T2V 14B WAN2.2] Hijabob Character Outfit

Hijabob v1

this lora stil experimental stage trying to figuring out how to tame wan2.2 t2v 14B specially the high lora. and the low lora can be used also for high lora. you can see the different result in each section.

Trigger word hijabob

example video on low is using low lora for both high and low.
and example video on high using both high and low lora.

example videos all contain metadata just download for example workflow, i generate with base Dyno high Wan2.2_T2V_A14B_4steps_25-09-28_Dyno_High_lightx2v-GGUF-Q_5_K_M
and base low Wan2.2-T2V-A14B-GGUF-Q_5_K_M

and only use low lightx2v lora low noise model
clip NSFW-Wan-UMT5-XXL-fp8

experimental lora weight
high : 2.0 ~ 3.0
low : 1.0 ~ 1.2

when combine with other lora, set other lora to
high : 0.7
low : 1.0


Training on local with rtx 3060 12gb vram and 48 gb ram. using Ai Toolkit Wan2.2 14B.
bit suprise it faster enough 1 night sleep only :3, around 7 ~ 8 hours.
transformer 4bit with ARA, rank 16, low vram,
optimizer Adamw8bit, linear
lr : 0,0001
wd : 0,0001

15 images 256px dataset train both high and low
3000 steps 1 batch.