Update README.md

README.md

@@ -11,12 +11,10 @@ tags:
 - pytorch
 - image-segmentation
 ---
-
 <div align="center">
 <img src="images/header.webp" width="800px" />
 
-Custom-trained models for face detection and segmentation across realistic, anime, and NSFW content.
-
+Custom-trained models for face detection and segmentation across realistic, anime, and stylized content.  
 Made for the **Forbidden Vision** ComfyUI custom nodes
 
 <a href="https://github.com/luxdelux7/ComfyUI-Forbidden-Vision">GitHub Repository</a>  
@@ -29,20 +27,20 @@ Made for the **Forbidden Vision** ComfyUI custom nodes
 
 ## 🎯 Why These Models Exist
 
-Traditional face models fail where it matters most for AI art workflows:
+Standard face detection models are optimized for narrow use cases and struggle in generative AI workflows. These models address four specific failure modes:
 
 | **Problem** | **Why It Matters** |
 |-------------|-------------------|
-| 🎨 **Domain-locked** | Existing models excel at *either* anime *or* realistic—never both |
-| 🔞 **NSFW blindness** | Most models trained only on SFW data break on adult content |
-| 👁️‍🗨️ **Detail blindness** | Most models miss anime eyebrows, real eyelashes etc. |
-| 🎲 **Generation artifacts** | Standard datasets don't include diffusion model quirks and failures |
+| 🎨 **Domain-locked** | Existing models excel at *either* anime *or* realistic — never both |
+| 🖼️ **Distribution mismatch** | Models trained on clean photography break on AI-generated imagery |
+| 👁️‍🗨️ **Detail blindness** | Most models miss stylized features like anime eyebrows, realistic eyelashes, etc. |
+| 🎲 **Generation artifacts** | Standard datasets don't include diffusion model quirks and failure modes |
 
 **These models solve all 4.**
 
 <div align="center">
 <img src="./images/masks.webp" alt="Mask Example" style="border-radius: 6px; box-shadow: 0 0 12px rgba(0,0,0,0.1);">
-<p><em>The segmentation model predicts face masks, stylistic eyebrows, eyelashes etc.</em></p>
+<p><em>The segmentation model predicts face masks including stylistic features like eyebrows and eyelashes.</em></p>
 </div>
 
 ---
@@ -51,7 +49,7 @@ Traditional face models fail where it matters most for AI art workflows:
 
 ### The Dataset Difference
 
-Built from **14,000+ manually annotated images** across the domains that actually matter for AI generation:
+Built from **14,000+ manually annotated images** spanning the full range of domains encountered in real generative AI workflows:
 
 <table>
 <tr>
@@ -61,7 +59,7 @@ Built from **14,000+ manually annotated images** across the domains that actuall
 - SDXL, SD1.5, Pony, Illustrious outputs
 - Curated Danbooru (anime styles)
 - Real photography
-- Full NSFW inclusion (no filtering)
+- Unfiltered image distributions across all content ratings
 
 </td>
 <td width="50%">
@@ -71,14 +69,13 @@ Built from **14,000+ manually annotated images** across the domains that actuall
 - ✓ Failed/broken generations
 - ✓ Low-quality artifacts
 - ✓ Unusual expressions & poses
-- ✓ Everything other models ignore
+- ✓ Edge cases other models ignore
 
 </td>
 </tr>
 </table>
 
 ### What This Means For You
-
 ```
 Traditional models: Trained on clean celebrity faces
          ↓
@@ -91,46 +88,55 @@ These models: Trained on what you actually generate
 
 **One model family. Every domain. Zero compromises.**
 
+---
+
 ## Model Details
 
 ### Face Detection (YOLOv11-Small)
 
-**Purpose:** Primary face detection with high recall
+**Purpose:** Primary face detection with high recall across mixed domains
 
 **Training Approach:**
-- After every training run, I ran the model on a new mixed dataset, hardmining failures and improving the dataset until an acceptable performance was reached
-- Trained at 640px resolution (inference should use same resolution)
+- Iterative hard-mining pipeline: after each training run, the model was evaluated on a new mixed dataset; failures were collected, corrected, and folded back into training until acceptable performance was reached
+- Trained at 640px resolution — inference should use the same resolution
 
 **Why YOLOv11-Small instead of nano?**  
-More reliable detection across mixed realistic/anime domains with acceptable speed tradeoff.
+More reliable detection across mixed realistic/anime domains with an acceptable speed tradeoff.
 
 ---
 
-
-### Segmentation (EfficientNet-v2)
+### Segmentation (EfficientNetV2-S)
 
 **Purpose:** Precise face mask generation
 
 **Training Approach:**
-- Dataset prepared using the Forbidden Vision YOLO model at 512px resolution
-- Iterative hardmine training in multiple phases:
-  - Train on the initial 700 samples
-  - Evaluate on remaining images to find failure cases
-  - Correct failed masks and add them to the dataset
-  - Retrain with the expanded dataset
-  - Repeat until failure cases drop to near-zero  
-    (final dataset: 4k+ images) 
+- Initial dataset prepared using the Forbidden Vision YOLO model at 512px resolution
+- Multi-phase iterative hard-mining:
+  1. Train on initial 700 samples
+  2. Evaluate on held-out images to surface failure cases
+  3. Correct failed masks and expand the dataset
+  4. Retrain on expanded dataset
+  5. Repeat until failure rate approaches zero
+  - Final dataset: 4,000+ images
 
 **Features:**
-- Detects and includes facial features other models ignore, like protruding anime eybrows, realistic eyelashes sticking out of the face etc.
-- Glasses and similar are treated as part of the face, even if sticking outside the face shape
-- NSFW friendly across both anime, realistic and 3d domains
+- Captures stylized facial features often missed by standard models: protruding anime eyebrows, realistic eyelashes extending beyond the face boundary, etc.
+- Treats accessories like glasses as part of the face region, even when they extend outside the face shape
+- Robust across anime, realistic, and 3D rendering styles — including content ratings that cause other models to fail
 
 ---
 
 ## Usage
 
-These models are automatically downloaded and used by the **Fixer** node in ComfyUI Forbidden Vision.
+These models are automatically downloaded and used by the **Fixer** node in ComfyUI Forbidden Vision. No manual setup required.
+
+---
+
+## Intended Use
+
+These models are designed for use in generative AI post-processing pipelines — specifically face detection and masking within ComfyUI workflows. They are not intended for surveillance, biometric identification, or any application involving real individuals without consent.
+
+---
 
 ## License