Arcane Style LoRA — MOSS-Image
English
What this is
Arcane Style is a style LoRA for MOSS-Image, trained to reproduce the painterly, high-contrast look of the Arcane animated series: thick oil-brush texture, hard-edged colour blocking, saturated violet / teal / amber palettes, and cinematic poster composition.
It is a style LoRA, not a character LoRA. It does not lock a face — use it on any subject and it paints them in this language.
Important: use it on the Base model
This LoRA was trained on MOSS-Image Base. It performs noticeably better on Base than on SFT — SFT has been aesthetic-tuned, and that post-training dilutes the LoRA's style signal. Train on Base, use it on Base.
Two ways to run it — and I recommend the fast one
There are two working configurations. Both give good results; the second is far faster and loses very little quality.
① Default — 50 steps, no speed LoRA. The straightforward path. Nothing extra to download.
② Recommended — 8 steps with the DMD speed LoRA. Pair this LoRA with the 8-step distillation of the Base model:
lodestones/MOSS-Image-Base-8-steps-dmd
It stacks cleanly on top of this LoRA. A 2.5 MP image renders in roughly 12 seconds, with only a very small quality difference — this is how every sample here was made.
Recommended parameters
Config ① — default (50 steps)
Base model: MOSS-Image Base (not SFT)
Sampler: euler
Scheduler: simple
Steps: 50
CFG: 3.0 – 4.0
Shift: 5.0
Config ② — recommended (8 steps, with the speed LoRA)
Base model: MOSS-Image Base (not SFT)
Sampler: euler
Scheduler: simple
Steps: 8
CFG: 1.0
Shift: 5.0
Speed LoRA: MOSS-Image-Base-8-steps-dmd at strength 1.0
Both configs
This LoRA's weight: 0.6 – 1.0 — 1.0 is the sweet spot; past ~1.2 colours start to over-saturate
Negative prompt:
worst quality, bad anatomy, watermark, jpeg artifactResolution: 1–2.5 MP works well. 16-pixel multiples only.
Aspect ratio
Pick by subject, the way you would for any real picture:
Portrait / character work → vertical, 3:4 (e.g. 1376×1840)
Environment / story / wide scene → horizontal, 16:9 (e.g. 2048×1216)
No trigger word
This is an unconditional style LoRA. There is no trigger phrase — just load it and adjust the weight. The style becomes the default rendering.
What it is good at
Cinematic character portraits with dramatic lighting
Poster-like graphic compositions and bold typography
Story beats: confrontation, tension, quiet moments
Environments and wide establishing shots
Chinese text rendering — MOSS-Image handles CJK typography unusually well, so Chinese poster work is a genuine strength here
How it was trained
168 images, captioned in plain natural language (content only, no style words — so the style is learned as the model's default, not as a prompt trigger)
MOSS-Image Base, rank 16 / alpha 16
50 epochs, learning rate 5e-5, AdamW, EMA off
Fixed clean-end share 40% to bias training toward layout and composition
Trained with the Fizgig trainer
Sample metadata
Every image uploaded here carries its full ComfyUI workflow and Civitai-compatible generation parameters embedded in the PNG. Hit Remix on any of them to load the exact graph that made it.
中文
这是什么
Arcane Style 是一个用于 MOSS-Image 的风格 LoRA，用来复刻《双城之战》那种厚涂油画质感、硬边色块、高饱和紫／青／琥珀色调，以及电影海报式的构图。
它是风格 LoRA，不是人物 LoRA。它不锁定某张脸——你可以用它画任何主体，它会用这套绘画语言去表现。
重要：请在 Base 模型上使用
这个 LoRA 是用 MOSS-Image Base 训练的，在 Base 上的表现明显好于 SFT。SFT 经过美学后训练，会稀释 LoRA 的风格信号。Base 训练，Base 使用。
两种跑法 —— 我更推荐快的那种
有两种可用的配置。效果都不错，第二种快得多，而画质损失很小。
① 默认 —— 50 步，不加速。 最直接的用法，不需要额外下载任何东西。
② 推荐 —— 8 步 + DMD 加速 LoRA。 搭配 Base 的 8 步蒸馏版：
lodestones/MOSS-Image-Base-8-steps-dmd
它与本 LoRA 可以干净地叠加。2.5 MP 的图约 12 秒一张，画质只有很小的差异——这里所有样图都是这么出的。
推荐参数
配置 ① —— 默认（50 步）
底模： MOSS-Image Base（不要用 SFT）
采样器： euler
调度器： simple
步数： 50
CFG： 3.0 – 4.0
Shift： 5.0
配置 ② —— 推荐（8 步 + 加速 LoRA）
底模： MOSS-Image Base（不要用 SFT）
采样器： euler
调度器： simple
步数： 8
CFG： 1.0
Shift： 5.0
加速 LoRA： MOSS-Image-Base-8-steps-dmd，权重 1.0
两种配置通用
本 LoRA 权重： 0.6 – 1.0，1.0 是甜点；超过约 1.2 色彩会开始过饱和
负向提示词：
worst quality, bad anatomy, watermark, jpeg artifact分辨率： 1–2.5 MP 都不错，边长必须是 16 的倍数
画幅建议
按题材选择，就像真实拍摄那样：
人像／角色 → 竖幅 3:4（如 1376×1840）
环境／故事／宽景 → 横幅 16:9（如 2048×1216）
没有触发词
这是一个无条件风格 LoRA。没有触发词——加载后直接调权重即可，风格就是默认渲染。
它擅长什么
电影感的角色肖像与戏剧性打光
海报式的图形化构图与粗体排版
故事性瞬间：对峙、紧张、静默的时刻
环境与宽阔的场景建立镜头
中文字体渲染——MOSS-Image 对中日韩文字的处理异常出色，中文海报是这个 LoRA 的一个真实优势
训练方式
168 张图，用自然语言打标（只描述内容、不含风格词——这样风格会被学成模型的默认渲染，而不是需要提示词触发的开关）
MOSS-Image Base，rank 16 / alpha 16
50 个 epoch，学习率 5e-5，AdamW，关闭 EMA
Clean-end share 固定 40%，让训练偏向布局与构图
使用 Fizgig 训练器
样图信息
这里上传的每张图都在 PNG 中内嵌了完整的 ComfyUI 工作流和 Civitai 兼容的生成参数。任意一张点 Remix 就能载入生成它的完整节点图。