SDXL LoRA weight : 0.7-0.9
Model : Animagine XL V3.1
Steps : 20-25
CFG Scale : 7
Sampler : Euler a
Prompt :
Kisugi Rui,1980s \(style\), film grain,retro artstyle, mature female, black hair,long hair,lipstick,(mole under mouth:0.8), blue eyes, purple bodysuit, bare shoulders, bodysuit,cleavage, breasts,high heels, cyan sash, sash, Description
FAQ
Comments (5)
太棒了，感谢大佬，一直在等猫眼三姐妹，自己不会调参数练的比较一般般
大佬，来生瞳和来生泪这两个MOD的高清分辨率参数选什么，有考虑用AutismmixPony作底模炼嘛
炼动画人物模型的话选普通的512x512，图片15 - 60张都行，调一下 batch = 2 , epochs = 20 ,
优化器默认ADW8BIT就行
一般 epochs 在12 - 20中间的LORA效果就可以了。
重要的是打标环节
Pony模型本身自带画风污染，对于还原画风有点不可控。目前有个模型页面有个BUG，等修复了之后，会炼Pony模型
resolution = "512,512"
min_bucket_reso = 256
max_bucket_reso = 2048
bucket_reso_steps = 64
save_every_n_epochs = 4
max_train_epochs = 20
train_batch_size = 2
learning_rate = 0.0001
unet_lr = 0.0001
text_encoder_lr = 0.00001
lr_scheduler = "cosine_with_restarts"
lr_warmup_steps = 0
lr_scheduler_num_cycles = 1
optimizer_type = "AdamW8bit"
network_module = "networks.lora"
network_dim = 8
network_alpha = 8
@jack_blue 非常感谢！
两位都对猫眼控贡献巨大！！猫眼的lora太少了。
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.