Sign In
homemodelsimagesvideospostsarticlesbountieschallengeseventsupdatesshop

Meowbah Character - Anima

Updated: Apr 21, 2026

characteranimevtuberwomananima

Download

1 variant available

SafeTensor

175.12 MB

Verified:

Type

LyCORIS

Stats

56

Reviews

(6)

Published

Mar 7, 2026

Base Model

Anima

Hash

AutoV2
30B978E773

Trigger Words

meowbah
pink neckerchief, neckerchief, pink skirt, pink sailor collar, pleated skirt, sailor collar, serafuku, single thighhigh, hair bow, school uniform, thighhighs, bow, pink bow, white shirt, long sleeves, asymmetrical legwear, red sailor collar, striped thighhighs, ribbon, black thighhighs
1girl, meowbah, pink eyes, cat ears, cat tail, tail bow, tail ribbon, tail jingle bell, cat girl, low twintails, multicolored hair, black hair, brown hair, pink hair, gradient hair, blunt bangs, long hair
white bra, wings, short twintails, underwear, white panties, frilled bra, underwear only, pink bow, thighhighs, angel wings, feathered wings, white wings, mini wings, small breasts, bare arms, lingerie, bridal garter, bare shoulders, frills
default creator card background decoration
luminarch's Avatar

luminarch

LyCORIS/loha

A lot more flexible than lora and trained with better dataset.

[network_arguments]
network_dim = 8
network_alpha = 4
network_module = "networks.loha"
network_train_unet_only = true
network_args = ["loraplus_unet_lr_ratio=2.0"]

[optimizer_arguments]
learning_rate = 1e-4
lr_scheduler = "cosine_with_restarts"
lr_scheduler_num_cycles = 5
lr_scheduler_power = 0
lr_warmup_steps = 0.1
optimizer_type = "came_pytorch.CAME"
optimizer_args = [ "weight_decay=0.01", "enable_cautious_update=True", "enable_cautious_weight_decay=True", "enable_stochastic_rounding=True", "enable_8bit=True"]

[training_arguments]
pretrained_model_name_or_path = ""
qwen3 = ""
vae = ""
max_train_epochs = 20
train_batch_size = 32
seed = 42
xformers = false
use_flash_attn = false
sdpa = true
lowram = false
no_half_vae = false
gradient_checkpointing = true
gradient_accumulation_steps = 1
max_data_loader_n_workers = 4
persistent_data_loader_workers = true
mixed_precision = "bf16"
full_bf16 = false
cache_latents = true
cache_latents_to_disk = true
cache_text_encoder_outputs = false

lora/locon

Trained on 19 images of Meowbah, no natural language was used. Can be used as a style lora too.

[general]
keep_tokens_separator = "|||"
shuffle_caption = true
flip_aug = false
caption_extension = ".txt"
enable_bucket = true
bucket_no_upscale = true
bucket_reso_steps = 32
min_bucket_reso = 288
max_bucket_reso = 2048

[[datasets]]
resolution = 768

[[datasets.subsets]]
caption_tag_dropout_rate = 0.1
num_repeats = 11
image_dir = ""
[network_arguments]
network_dim = 64
network_alpha = 32
network_module = "networks.lora_anima"
network_train_unet_only = true

[optimizer_arguments]
learning_rate = 4e-4
lr_scheduler = "cosine_with_restarts"
lr_scheduler_num_cycles = 3
lr_scheduler_power = 0
lr_warmup_steps = 0.1
optimizer_type = "came_pytorch.CAME"
optimizer_args = [ "weight_decay=0.01", "enable_cautious_update=True", "enable_cautious_weight_decay=True", "enable_stochastic_rounding=True", "enable_8bit=True"]

[training_arguments]
pretrained_model_name_or_path = ""
qwen3 = ""
vae = ""
max_train_epochs = 15
train_batch_size = 32
seed = 42
xformers = false
use_flash_attn = false
sdpa = true
lowram = false
no_half_vae = false
gradient_checkpointing = true
gradient_accumulation_steps = 1
max_data_loader_n_workers = 4
persistent_data_loader_workers = true
mixed_precision = "bf16"
full_bf16 = false
cache_latents = true
cache_latents_to_disk = true
cache_text_encoder_outputs = false

[sampling]
sample_every_n_epochs = 1
sample_prompts = ""
sample_sampler = "euler_a"
sample_at_first = true

[saving_arguments]
save_precision = "bf16"
save_model_as = "safetensors"
save_every_n_epochs = 1
save_last_n_epochs = 7
output_name = ""
output_dir = ""
log_prefix = ""
logging_dir = ""
wandb_run_name = ""
wandb_api_key = ""
log_with = "wandb"

  • Sampler: euler/er_sde/euler_ancestral

  • CFG: 5.0

  • Steps: 20

  • weight: 1.0