### About the Model

This is a high-fidelity 3D CGI and cinematic game-render style LoRA trained specifically for the Anima model architecture.

It infuses generations with a distinct, clean 3D-engine aesthetic reminiscent of modern offline raytraced renders or advanced real-time game engines (such as Capcom's RE Engine or Unreal Engine 5). It focuses on capturing micro-details: realistic sub-surface scattering on skin, volumetric shadows, cinematic lighting, and sharp textures on fabrics and hair, all while maintaining the core anatomical structure of your target characters.

---

### How to Use & Prompting Strategy

To get the most out of this style without overwriting character concepts, use the following approach:

* Primary Trigger Word: @lllsuperlllcgi

(Note: The @ prefix is highly optimized for Anima's Qwen3 text encoder, allowing it to isolate the render style from the subject's physical traits).

* Style Anchors (Optional but Recommended):

Including tags like cgi or (blender \(medium\):0.7) helps steer the model towards that distinct CGI output, balancing the digital-art feel with realism.

* Tagging Tip: This LoRA was trained with UNet-only weights to preserve characters. Keep your prompts descriptive of what is in the scene (clothing, background, hair) so the LoRA can focus strictly on how it is rendered.

---

### Recommended Generation Settings

For optimal results, configure your UI with the following parameters:

* LoRA Weight: 0.7 to 0.9, 0.9 is recommended for a highly pronounced 3D render effect).

* Base Model: Anima v1.0 Base (or fine-tunes such as waiANIMA_v10Base10).

* Sampler: DPM2 (or Euler)

* Schedule Type: Karras (or Normal)

* Steps: 25 – 35

* CFG Scale: 1.5 – 3.5 (Crucial: keep CFG low; a value of 3.0 is the sweet spot to prevent color frying on Flow-based models).

* Discrete Flow Shift: 3.0 (or 1.0 depending on your scheduler setup).

* Resolution: Optimized for standard high-resolution outputs like 832x1152.

---

### 💡 Pro-Tip: Enhance the CGI Realism with "Amateur Filter"

To push the 3D-CGI output to the next level and get rid of the typical plastic "AI sheen," we highly recommend using the Amateur Filter extension (available for WebUI / Forge) with these exact settings:

* Sharpen Intensity: 0.01

* Desaturation Factor: 0.85 (helps blend colors and mimics cinematic grading)

* Grain Amount: 0.02 (adds subtle digital sensor noise)

* JPEG Quality: 100

#### Example Prompt:

> <lora:lllsuperlllCGI_style_FINAL:0.9> @lllsuperlllcgi, (blender \(medium\):0.7), cgi, 1girl, 2b \(nier:automata\), (veins, sweat:0.7), very pale skin, black robe, blindfold covering eyes, on side, medium breasts

#### Recommended Negative Prompt:

> worst quality, low quality, score_1, score_2, score_3, artist name, blurry, jpeg artifacts, lowres, censor, text speach, (bad fingers, 3 fingers, 4 fingers:1.2), (multiple view:1.2), grayscale, black and white, neon colors, (names, artists name:1.1), (watermarks, signature:1.2), censored, wrinkled skin, bad hands, bad legs, speech bubble, sticker