Not loaded
Download
1 variant available

42
233
License:
AnimaThe Anima Model is licensed by CircleStone Labs LLC. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
Built on NVIDIA Cosmos
Introduction
This is a SeiroA style LoRA trained on Anima Base 1.0. SeiroA is an artist who primarily creates adult manga.
I trained this LoRA at a friend's request. Along the way, I also learned some data-curation strategies for noisy training material such as manga, so I may release more models based on other manga artists whose work I personally like!
Usage
Recommended LoRA strength: 1.0
Trigger word: @seiroa
Training Details
Base model: Anima Base 1.0
Trainer:
sorryhyun/anima_loraDataset: 285 character crops from manga
LoRA rank / alpha:
32/128Learning rate:
2e-5Optimizer: AdamW
Scheduler: Cosine, with 5% warmup
Batch size:
1per GPUTraining duration: 20 epochs / 1,900 steps
Precision: BF16
Resolution: Aspect-ratio buckets at approximately 1 megapixel
GPU: 3 × NVIDIA RTX 3090
简介
这是基于 Anima Base 1.0 训练的 SeiroA 画风 LoRA。SeiroA 是一位主要绘制成人漫画的画师。
这个 LoRA 应朋友的请求训练。我在训练中也学习了一些数据清洗的策略(针对漫画之类的噪音较大的训练数据),所以我接下来可能还会发布一些其他我觉得较好的漫画画师的模型作品!
使用
推荐 LoRA 强度: 1.0
激活词: @seiroa
训练参数
基础模型: Anima Base 1.0
训练工具:
sorryhyun/anima_lora数据集: 285 张人物裁剪(来自漫画)
LoRA rank / alpha:
32/128学习率:
2e-5优化器: AdamW
学习率调度: Cosine,5% warmup
Batch size: 每卡 Batch 为 1
训练量: 20 epochs / 1,900 steps
训练精度: BF16
训练分辨率: 约 1 M 像素的宽高比分桶
显卡: 3 × NVIDIA RTX 3090



