Hey everyone, here's a character model of Wanwan from Onlinepetclub. I came across her design and really liked it, so I decided to create this.
I know I haven't been very active on Civitai or given much Buzz lately. Life's been pretty busy with work, and I haven't had a lot of free time. If you like what I do, any support you can show would be a huge motivation for me to keep creating!
Regarding the dataset, I spent a lot of time carefully selecting and filtering the images to get the best possible quality. I have to say, the results turned out even better than I expected.
I have a lot of characters planned for the future, so I haven't decided who's next just yet. The biggest challenge is that some characters have very few images available, which means I'd need to do a lot of manual editing to create a complete dataset.
Anyway, here's the training info. Hope you enjoy it!
Training Information:
Epochs: 9
Batch Size: 1
Resolution: 1024px
Enable Bucket: True
Image Count: 18
Tags:
base
w4nw4n, 1girl, solo, dog girl, animal ears, dog ears, short hair, brown hair, light brown hair, hair between eyes, black eyes, ahoge, dog tail, blush stickers, outfit
school uniform, serafuku, white shirt, shirt, blue sailor collar, sailor collar, red neckerchief, collarbone,long sleeves, loafers, blue skirt, pleated skirt,black thighhighs, zettai ryouiki, ```Description
### Training Information (Anima DiT)
+ Output Name: bf16_Anima_Char_Wanwan_Retrain
+ Network Module: networks.lora_anima
+ Epochs: 12
+ Max Train Steps: Auto (Epoch)
+ Learning Rate: 0.0001
+ Text Encoder LR: 5e-05
+ Optimizer: AdamW8bit
+ Mixed Precision: bf16
+ LR Scheduler: constant
+ LR Warmup Steps: 0
+ Rank: 16/16 (Dim, Alpha)
### Anima Arguments:
+ timestep_sample_method: logit_normal
+ discrete_flow_shift: 3
+ weighting_scheme: logit_normal
### Dataset Info (dataset.toml):
+ Batch Size: 2
+ Resolution: 1536x1536
+ Steps (calculated): 528
+ Bucketing: True
### Datasets Parameter:
+ Anime : images: 13 | repeats: 4 | keep_tokens: 1 | flip_aug: False | shuffle_caption: False
+ 2_chibi : images: 5 | repeats: 1 | keep_tokens: 1 | flip_aug: False | shuffle_caption: False
+ Anime. no agoe : images: 5 | repeats: 4 | keep_tokens: 1 | flip_aug: False | shuffle_caption: False
+ Chibi no agoe : images: 7 | repeats: 1 | keep_tokens: 1 | flip_aug: False | shuffle_caption: False
+ Anime new : images: 2 | repeats: 2 | keep_tokens: 1 | flip_aug: False | shuffle_caption: False