CoolFace
Modelpublic

SAkizuki/Anima-yoshinari-ko

sourceHugging Facemitupdated 4mo agoView on Hugging Face
0likes26downloads
Model Card

Anima-yoshinari-ko

Anima 兼容的画风 LoKR,旨在复现动画师吉成鋼在《向山进发4》ED 中的手绘水彩淡彩风格。

模型简介

本模型是一个基于 Anima 训练的 LoKR(Low-Rank Kronecker Product) 画风适配器,目标是捕捉《向山进发4》(Yama no Susume: Next Summit)BD 版 Non-Credit ED 动画截图中呈现的吉成鋼画风特征——手绘质感、水彩渲染、淡雅配色与柔和线条。

本模型使用 Anima Base v1.0 作为底模,训练集约 300 张来自《向山进发4》BD NCED 的动画截图,经 14 轮迭代训练而成。

触发词和推荐强度

请使用@yoshinari ko触发此画风,强度建议设置为0.5~1.5,过强的模型强度会对构图产生破坏。

推荐用法

示例图像

masterpiece, best quality, very aesthetic,@yoshinari ko,1girl,close-up, white_hair, blue eyes, medium hair, high ponytail, sidelocks, headset, 

masterpiece, best quality, score_7, safe, 2girls, yukimura aoi, yama no susume, grey hair, short twintails, green eyes, hairclip, jacket, backpack, gloves, hiking pole, boots, 
@yoshinari ko
kuraue hinata, black hair, short twintails, purple eyes, hairclip, jacket, backpack, gloves, hiking pole, boots, 
smile, excited, facing viewer, looking at viewer, holding hands, helping, winter, mountain, snow, climbing, no lineart, watercolor effect, painterly, brush stroke, full body

Both girls stand close together on a snowy mountain slope, holding hands and supporting each other up the climb while facing the viewer with bright smiles and excited expressions. The full-body watercolor painting shows no lineart and visible brush strokes across the winter landscape. Soft diffused light illuminates the characters against a pale, snowy sky, keeping them as the central focus in a near-foreground composition.
masterpiece, best quality, very aesthetic

masterpiece, best quality, score_7, safe, 1girl, solo, (blonde hair:2), blonde eyes, hair between eyes, short hair, ahoge, twintails, short twintails, low twintails, sidelocks, chinese clothes, hanfu, short hanfu, (jiaoling ruqun:2), banbi, short skirt, short mamianqun, sash, red sash, beizi \(clothing\), fingerless gloves, hairclip, tassel, converse, hand on own cheek, seductive smile, looking at viewer, portrait, simple background, monochrome background
@yoshinari ko

A blonde girl with short twintails and an ahoge leans slightly forward, one hand resting lightly on her cheek as she gazes directly at the viewer with a confident, mischievous smile. Her bright eyes sparkle with playful charm, framed by sidelocks and hair between eyes. Soft indoor lighting illuminates her Chinese hanfu outfit against a clean, light-toned background, creating a warm and intimate portrait composition.
masterpiece, best quality, very aesthetic

模型规格

项目详情
模型类型LoKR(Low-Rank Kronecker Product)
底模Anima Base v1.0
LoKR Rank64
LoKR Alpha64
LoKR Factor8
Full Matrix是
训练精度FP16
输出格式SafeTensors

训练详情

项目详情
训练轮数14 epochs
批大小2
梯度累积2
优化器Automagic
学习率5×10⁻⁵(UNet)
学习率调度Cosine with Restarts(3 个周期)
预热步数100
训练分辨率1024×576
最小/最大 Bucket256 / 2048
CLIP Skip2
扩散偏移3(Shift)

训练数据

  • —来源: 《向山进发4》(Yama no Susume: Next Summit)BD 版
  • —内容: Non-Credit Ending 动画截图
  • —数量: 约 300 张
  • —触发词标识: @yoshinari ko

画风特点

吉成鋼的画风在本模型中主要体现为以下特征:

  • —手绘质感: 线条保留手绘的不规则感与温度,避免数码绘的过于干净与锐利
  • —水彩渲染: 色彩以水彩方式晕染过渡,色彩层次丰富但不浓烈
  • —淡彩配色: 整体色调偏淡雅,以低饱和度的柔和色彩为主
  • —柔和线条: 轮廓线较细且有轻微的笔触变化,呈现出铅笔或色铅笔的质感
  • —光影处理: 光影过渡柔和自然,避免硬边阴影

兼容性

本模型基于 Anima Base v1.0 训练,理论兼容以下环境:

  • —ComfyUI(使用LoRA加载器(仅模型)加载)
  • —支持 LoKR 加载的 Stable Diffusion 工具链
  • —所有支持 SafeTensors 格式的 SD 工具