This document is bilingual in Chinese and English. English users please read below
Airy Photography Style | 空気感写真style
一个用于 MOSS-Image 的空气感写真风格 LoRA。主要想让画面更明亮、肤色更舒服，同时稳定在真实人物、Cosplay 写真的方向。
整体偏通透、柔和的写真观感，适合人像、日常生活照和 Cosplay。下面这些是我目前测试下来的个人感受，欢迎大家尝试不同场景。
两个版本
3200 steps｜默认版本：目前作为默认发布版本，建议从这一版开始体验。
1360 steps｜更强风格：在我的测试中，这一版的风格感反而更强。喜欢更明显的写真处理，可以试试这个版本。
左边=无lora，中间=1360 steps ， 右边3200 steps
步数更多不一定代表风格更强，也不一定每个场景都更好，两个版本可以按喜好选择。
目前观察到的优点
提亮画面：让画面更明亮，带来比较明显的通透感。
改善肤色：肤色更柔和、白皙，偏向经过整理的写真效果。
锚定写真风格：尤其是生成动漫角色 Cosplay 时，更容易保持真人写真的方向，减少向二次元插画偏移的情况。
一定程度上提升颜值：脸部更精致，妆面和整体观感更接近 Cosplay 写真。
部分场景中减少肢体崩坏：这是个人对比测试中的感受，并不是稳定的人体修复功能。
使用时需要注意
MOSS 底模本身仍然会出现手指、四肢和身体结构崩坏，需要多抽卡。 加载这个 LoRA 后也不能保证解决，复杂动作尤其如此。
它也不只影响颜色和肤质。在部分测试中，会改变取景距离、构图，以及手、道具和身体之间的关系；夸张表情、复杂视角或带有蓄势感的动作，有时会被拉回更规整的写真姿势。
部分发型、脸型也可能受到素材的影响。不过这些影响并不是每张都会出现，具体取决于提示词、种子和 LoRA 权重。
使用参考
适用模型：MOSS-Image
测试底模：MOSS-Image SFT
使用4.5cfg 25步来生成图片以获得速度与质量的平衡。
触发词(没有也生效)：
kuukiikan_style展示测试主要使用 1.0 权重；如果觉得风格干预过强，可以降低权重再比较。
Airy Photography Style | 空気感写真style
An airy photographic style LoRA for MOSS-Image, designed to brighten images, soften skin tones, and keep generations closer to real photography and cosplay photoshoots.
The overall look is bright, soft, and polished. It works well for portraits, everyday scenes, and cosplay photography. The observations below come from my own testing—results will vary by prompt, seed, and LoRA strength.
Versions
3200 steps | Default: The default release and a good place to start.
1360 steps | Stronger style: In my tests, this earlier checkpoint actually produces a stronger stylistic effect. Try it if you prefer a more pronounced photoshoot look.
Left=no lora, middle=1360 steps, right=3200 steps
More training steps do not necessarily mean a stronger style or better results in every scene. Choose whichever version suits your taste.
What it does well
Brighter images: Adds a brighter, airier appearance.
Softer skin tones: Tends toward fairer, softer skin with a polished photographic finish.
More consistent photographic styling: Particularly useful for anime character cosplay, helping results stay closer to real cosplay photography instead of drifting toward illustration.
More flattering faces: Can produce more refined facial features and a more polished makeup look.
Fewer anatomy errors in some cases: I noticed improvements in some comparisons, but this is a personal observation—not a reliable anatomy correction feature.
Limitations
MOSS itself can still struggle with hands, limbs, and body structure. Expect to generate multiple images and select the best results. This LoRA does not eliminate those issues, especially with complex poses.
It also affects more than color and skin rendering. In some tests, it changed framing, camera distance, and the relationships between hands, props, and the body. Exaggerated expressions, unusual angles, or poses suggesting imminent movement can become more conventional photoshoot poses.
Hairstyles and facial features may also pick up tendencies from the training images. These effects vary across prompts and seeds.
Usage
Model family: MOSS-Image
Tested base: MOSS-Image SFT
Use 4.5cfg with 25 steps to generate images to achieve a balance between speed and quality.
Trigger（Not even effective）:
kuukiikan_styleMost comparison images were generated at 1.0 strength. Lower the strength if the style interferes too much with your intended composition.