Update README.md

README.md

@@ -13,14 +13,41 @@ license: mit
 
 # 🚀 SAM2 + MatAnyone Video Background Changer
 
-This Hugging Face Space provides a powerful video background replacement tool using state-of-the-art AI models:
+A Hugging Face Space for high-quality **video background replacement** using cutting-edge AI models:
 
-- **Segmentation**: `Segment Anything 2 (SAM2)` for highly accurate foreground detection.
-- **Matting**: `MatAnyone` for professional, cinema-quality video matting.
+- 🦾 **Segmentation:** [Segment Anything 2 (SAM2)](https://github.com/facebookresearch/segment-anything-2) for precise, state-of-the-art foreground detection.
+- 🎬 **Matting:** [MatAnyone](https://github.com/pq-yang/MatAnyone) for cinema-quality subject/background compositing.
 
-## Key Features
+---
+
+## ✨ Key Features
+
+- 🎞️ **Cinema-Quality Two-Stage Processing:**  
+  Original → Green Screen → Final Background, for ultra-clean edge separation and realistic results.
+- 🖼️ **4 Background Methods:**  
+  - 📷 Upload custom images
+  - 🎨 Use professional presets
+  - 🌈 Create custom gradients
+  - 🤖 Generate backgrounds with AI prompts
+- ⚡ **Lazy Model Loading & GPU Optimization:**  
+  Efficient use of the T4 GPU — models are loaded only when needed.
+- 🖥️ **Intuitive Gradio Interface:**  
+  Just upload your video, pick a background, and hit process!
+- 🧹 **Automatic Memory Management:**  
+  Models and GPU cache are cleared after each run for maximum reliability.
+
+---
+
+## ⚙️ Configuration Notice
+
+> **Required Model Configs:**  
+> This app requires one or more [SAM2 model config YAML files](https://github.com/facebookresearch/segment-anything-2/tree/main/configs) to be present in a `configs/` directory at the root of your repo.
+>
+> **Examples:**  
+> - `configs/sam2_hiera_large.yaml` (**recommended**)  
+> - `configs/sam2_hiera_tiny.yaml` (for CPU/fallback)
+>
+> If missing, download from the official SAM2 repo and add them to `configs/`.
+
+### 📁 Example Directory Structure
 
-- **Lazy Loading**: Models are loaded into GPU memory only when needed, keeping the Space resource-efficient.
-- **GPU Optimization**: The Space is configured to use a T4 Small GPU for a balance of performance and availability.
-- **Gradio Interface**: A simple and intuitive user interface for uploading videos and background images.
-- **Memory Management**: After each run, models are offloaded and the GPU cache is cleared to minimize costs.