Sign In

浮生云绘 style | MOSS Image LoRA

Updated: Nov 16, 2026

style

Download

1 variant available

bf16 SafeTensor

no83_dongfang_r32_a128_1mp_e140.safetensors

BF16, good balance • 189.05 MB

Verified:

Type
LoRA
Stats

85

Reviews
Published

Nov 16, 2026

Base Model
Training
Steps: 4,760Epochs: 140
Usage Tips
Strength: 1
Hash
AutoV2C381F71BBA
Trigger Words
@no83_dongfang
01_钟影回廊_00001_.png

训练素材来自 B 站「浮生云绘」制作的画集。这款 LoRA 主要学习画面的构图和光影安排,包括主体落点、前后景层次,以及如何用光线和色彩把视线引向主体。

e140 是这次训练中美学表现最强的版本,在画面层次、光色和装饰细节上表现最好。在我的测试中,随着训练次数增加,复杂手部动作也更容易出错,尤其是双手交叠、捏持细小道具时。如果画面的重点是手势或手与道具的接触,可以先试 e110。

适用底模为 MOSS Image Base,触发词为 @no83_dongfang。

English Version

Dongfang · Composition & Lighting | MOSS Image LoRA

Trained on images from an art collection put together by 浮生云绘 on Bilibili. This LoRA focuses on composition and lighting, including subject placement, depth between foreground and background, and how light and color guide the eye through a scene.

e140 gives the strongest aesthetic results of this training run, particularly in depth, lighting, color, and ornamental detail. In my tests, complex hand poses became more error-prone at higher training counts, especially when hands overlap or fingers grip small objects. If the image depends on a precise gesture or grip, e110 is a good starting point.

Use with MOSS Image Base. The trigger word is @no83_dongfang.