# 📸 Photo Anime — 2.5D Korean Manhwa Semi-Realism LoRA
**Created by AniBoy** | Trained on **Illustrious-XL v0.1**
Tired of flat 2D anime drawings with plastic, sticker-like skin, or hyper-realistic models that erase all anime facial charm?
**Photo Anime** is trained on a meticulously curated, noise-free dataset of high-end 2.5D manhwa artwork. It bridges the gap between stylized anime aesthetics and real-world photographic rendering: **tangible skin pores, porcelain translucency, subsurface scattering, 3D volumetric ambient occlusion, and micro-fabric textures** (ribbed knits, leather reflections, silk drape) — while completely suppressing harsh cartoon black lineart and flat cel-shading.
---
## ⚡ Quick Start & Trigger Word
* **Trigger Word**: photo_anime (Place at the very beginning of your prompt)
* **LoRA Tag**: <lora:photo_anime_style_v1.0:1.0>
* **Recommended Weight**: 0.95 – 1.05 (Use 1.0 on stubborn 2D anime checkpoints like NoobAI to overpower flat cel-shading; 0.75 – 0.85 on native Illustrious-XL).
---
## ⚙️ Best Generation Settings
| Setting | Recommended Value | Notes |
| :--- | :--- | :--- |
| **Base Model** | Illustrious-XL v0.1 / NoobAI / SDXL Anime | Tested thoroughly on NoobAI & Illustrious |
| **Sampler** | Euler or Euler a | Extremely clean skin gradients |
| **Schedule Type** | Karras or Automatic | Smooth step progression |
| **Steps** | 28 – 32 | Sweet spot for micro-detail |
| **CFG Scale** | 4.5 – 5.0 | **Do NOT exceed 6.0** (high CFG enforces cartoon lines) |
| **Base Resolution** | 1024 × 1024 or 832 × 1216 | SDXL native aspect ratios |
| **Hires. Fix Upscaler** | *4x-UltraSharp** or R-ESRGAN 4x+ | **Crucial! Read upscaler warning below** |
| **Hires Denoising** | 0.32 – 0.38 | Upscale by 1.5x, 18–20 steps |
---
## ⚠️ CRITICAL USAGE ADVISORY (Read Before Generating!)
### 1. 🚨 Male Characters & Full-Body Shots (USE ADETAILER)
> [!WARNING]
> **Dataset Transparency**: This model was trained on 57 high-fidelity images, primarily focused on female subjects (49 female / 8 male) and cowboy/bust compositions (only 4 full-body shots).
>
> * **For Female Bust / Cowboy Shots**: Keep **ADetailer OFF**. At close-to-medium range, the LoRA natively renders stunning 3D skin pores and lip volume. Running ADetailer with a 2D base checkpoint will actually erase the skin texture!
> * **For Male Characters 1boy)**: **ENABLE ADetailer** face_yolov8s.pt, denoise 0.30–0.35). Because male data was limited in the training set, ADetailer prevents facial distortion on male characters.
> * **For Full-Body Shots**: **ENABLE ADetailer** face_yolov8s.pt, denoise 0.30–0.35) to maintain facial sharpness at small pixel counts.
### 2. 🚫 The Upscaler Trap (Never use Anime Upscalers)
❌ *DO NOT USE**: RealESRGAN_x4plus_anime_6B or any other anime-specific upscaler. Anime upscalers are programmed to treat skin pores and subtle gradients as "noise" and smooth them out into flat vector color fills!
✅ *ALWAYS USE**: 4x-UltraSharp, R-ESRGAN 4x+, or Nomos8k. These preserve skin pores, peach fuzz, and realistic fabric grain.
---
## 🧩 Stacking with Other LoRAs
This LoRA stacks phenomenally with character LoRAs and other styles due to clean training:
* **Character LoRA + Photo Anime**:
Character LoRA: 0.75 – 0.85 (Locks identity, hair, and costume)*
Photo Anime: 0.65 – 0.75 (Applies 3D skin, realistic lighting, and fabric)*
* Example: <lora:your_character:0.8>, <lora:photo_anime_style_v1.0:0.7>, photo_anime, [character triggers]...
* **Style + Style (e.g., Cultivation / Xianxia / Cyberpunk)**:
* Style LoRA: 0.60 – 0.65
* Photo Anime: 0.70 – 0.75
* Keep the combined weight sum under 1.40 to avoid overbaking.