# 🌟 3DCG Master: Stylized Anime 3D & Skin Gloss LoRA
> 🔥 Latest Version: V3 (Pure Anime 3D & Style Baking) — Recommended for all new generations.
> Seamless prompt compatibility, pure Japanese anime facial aesthetics, zero lighting conflicts.
---
## 🚀 [V3.0] Pure Style Baking & Anime 3D (Latest & Recommended)
### 🌟 What makes V3 a Game Changer?
In previous versions, 3D LoRAs relied heavily on tags like 3d render, cgi, or cinematic lighting. This caused severe lighting conflicts whenever users requested dim light, soft daylight, or film atmospheres.
In V3, all render and lighting words were scrubbed from training captions (Style Baking).
The physical 3D subsurface scattering (SSS), glossy highlights, and volumetric depth are now baked directly into the model's neural latent space.
- Zero Lighting Conflicts: Works flawlessly with dim lighting, candid 35mm flash, golden hour, or moody shadows.
- Pure Japanese Anime Face: 47.5% of the dataset (358 / 753 images) is concentrated on the beloved trio (**M71, cakiiBB, HydraFXX**). Western game model drift is 100% eliminated.
- Alpha 48.0 (1.5x scale): High fidelity, razor-sharp outlines, and zero color artifacting even at weight 1.0.
### ⚙️ V3 Recommended Generation Settings
- Base Model: MOSS Image Base mossImage_base)
- LoRA Weight: 0.8 ~ 1.0 (1.0 recommended)
- Trigger Word: style_3dcg_master
- Sampler:
- euler: Pure porcelain doll skin, clean anime figure look (zero noise).
- er_sde: Maximum subsurface micro-detail, shiny wet sheen, sharp individual hair strands.
- Scheduler: simple
- Steps: 40 ~ 50
- CFG Scale: 2.8 ~ 3.5 (Optimal: 3.0)
### 📝 V3 Prompting Examples
- Clean Showcase:
best quality, highres, style_3dcg_master, 1girl, solo, cute smile, looking at viewer, soft daylight
- Moody Rim Light / Character Showcase:
best quality, highres, style_3dcg_master, makima (chainsaw man), 1girl, red hair, braided ponytail, yellow eyes, ringed eyes, white collared shirt, indoor, dim moody rim lighting, highly detailed face, glossy lips
---
## 📦 [V1.0] Original Release (Archived / Legacy)
Trained on a curated, high-resolution dataset of 225 images synthesized from 5 legendary 3DCG masters (GeneralButch, Fugtrup, M71, cakiiBB, and Hydrafxx) with 2,250 steps.
This model captures the best elements of all five styles:
- Specular Wet Gloss: Stunning liquid sheen and mirror reflections on skin.
- Subsurface Scattering (SSS): Warm, translucent, and realistic flesh tones without plastic stiffness.
- Smooth Anatomy: Flawless body contours, glute curves, and soft realistic skin folds under tension.
- Micro Details: Pristine fingers, nails, and environment reflections.
### ⚙️ V1 Recommended Settings
- Base Model: MOSS-Image mossImage_sft_pruned_int8.safetensors)
- LoRA Weight: 0.8 ~ 1.0 (Recommended: 1.0 for full punchy 3DCG aesthetic)
- Trigger Words: style_3dcg_master, 3d, 3d (artwork)
### 💡 V1 Prompt Suggestions
- Add positive anchors: subsurface scattering, sharp specular highlights, high polygon 3d model render, stylized 3dcg shading
- Works seamlessly on both SFW portraits and explicit NSFW anatomy!
Description
# 3DCG Master - Pure Anime 3D & Style Baking (V3)
The definitive Japanese Anime 3DCG & SSS Figure LoRA for MOSS Image Base.
Trained on 753 curated images with a massive 47.5% focus on M71, cakiiBB, and HydraFXX.
---
### 🌟 What makes V3 a game changer? (Pure Style Baking)
In previous versions, 3D LoRAs relied heavily on tags like 3d render, cgi, or cinematic lighting. This caused severe lighting conflicts when users wanted dim light, soft daylight, or film atmospheres.
In V3, all render and lighting words were scrubbed from training captions (Style Baking).
The physical 3D subsurface scattering (SSS), glossy highlights, and volumetric depth are now baked directly into the model's neural latent space.
- No lighting conflicts: Works flawlessly with dim lighting, candid 35mm flash, golden hour, or moody shadows.
- Pure Japanese Anime Face: 47.5% of the dataset is concentrated on the beloved trio (M71, cakiiBB, HydraFXX). Western game model drift is 100% eliminated.
- Alpha 48.0 (1.5x scale): High fidelity, razor-sharp outlines, and zero color artifacting even at weight 1.0.
---
### ⚙️ Recommended Generation Settings
- Base Model: MOSS Image Base mossImage_base)
- LoRA Weight: 0.8 ~ 1.0 (1.0 recommended)
- Trigger Word: style_3dcg_master
### 📝 Prompting Examples
Clean Showcase:
> best quality, highres, style_3dcg_master, 1girl, solo, cute smile, looking at viewer, soft daylight
Moody Rim Light / Character Showcase:
> best quality, highres, style_3dcg_master, makima (chainsaw man), 1girl, red hair, braided ponytail, yellow eyes, ringed eyes, white collared shirt, indoor, dim moody rim lighting, highly detailed face, glossy lips
Negative Prompt:
> worst quality, low quality, bad anatomy, deformed hands, missing fingers, extra limbs, watermark, text, signature, blurry, oversaturated
---
### 📊 Evolution (V1 ➔ V2 ➔ V3)
- V1 (225 imgs): Prototype. Prone to lighting prompt clashes and western face drift.
- V2 (570 imgs): Expanded anatomy and poses, but still suffered from tag conflicts.
- V3 (753 imgs): True Style Baking + 47.5% Japanese Trio focus + Alpha 48.0. Zero tag pollution, pure anime aesthetics.