文件清单 线路：镜像（hf-mirror.com） · 切换
下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。
说明
3D Realistic Portrait Enhancer
作者：baowy
3D Realistic Portrait Enhancer · 3D 写实人像增强
A 3D-realistic portrait enhancer for MOSS-Image. Strengthens subsurface scattering, skin texture and facial volume while keeping the underlying structure intact — no face reshaping, no anime drift.
What it does
At higher strength the render picks up:
- Subsurface scattering — skin reads as translucent rather than painted
- Facial volume — cheekbones, jawline and nose bridge get real depth
- Skin texture — pore-level detail and fine tonal variation
- Tighter light falloff — highlights concentrate, shadows separate
All four happen together, which is why it reads as "3D rendered" rather than "filtered". The face shape and proportions stay stable across the whole range.
Recommended strength
| Strength | Result |
|---|---|
| 0.8 – 1.0 | Subtle. More depth and texture, still close to the base render. |
| 1.5 | Clear 3D render feel. Best balance of detail and realism. (recommended) |
| 1.8 | Maximum volume and skin translucency. Slight face refinement. |
| 2.0 | Upper limit — skin starts to look waxy. |
Use 1.0 – 1.5 for most work. There is no negative direction: this is a fixed blend, not a slider, and negative values will degrade the render.
What it works on
Measured on identical seeds, difference from base at strength 1.8:
| Subject | Effect |
|---|---|
| Portrait / half body | Strong — skin, volume, light separation |
| Full body | Strong — fabric weight and body structure also improve |
| Male subjects | Strong — no feminization, structure stays masculine |
| Objects / macro | Weak — slight material benefit only |
| Scenery / interiors | Almost none — do not use it for landscapes |
It is a portrait tool. On scenes without people it does close to nothing — that also means it will not pollute your background.
Settings
MOSS_8-steps-dmd2-lora 1.0 MOSS_3DReal_Portrait 1.0 - 1.5 Steps 8 CFG 1.0 Sampler euler Scheduler simple Resolution 1024x1360 (any multiple of 16)
No trigger words required. Works with any prompt.
Samples
All comparison images used the same seed (552314880), strength 0 / 1.0 / 1.5 / 1.8, 1024×1360.## 中文说明
MOSS-Image 的 3D 写实人像增强 LoRA。提高强度会同时加强三件事： 次表面散射（皮肤通透感）、面部体积（颧骨、下颌、鼻梁的立体度）、 皮肤纹理（毛孔级细节）。
推荐 1.0 – 1.5，1.8 是立体感最强的一档，2.0 皮肤开始有蜡感。
这不是滑块，没有负方向，负值会劣化画面。
适用范围：人像、半身、全身、男性都有效，男性不会被女性化。 不适用：纯场景、风景、室内空镜——实测几乎没有效果，别浪费显存。