Update README.md

README.md

@@ -4,11 +4,55 @@ emoji: 🎬
 colorFrom: red
 colorTo: purple
 sdk: gradio
+sdk_version: 6.18.0
+app_file: app.py
+pinned: false
+license: apache-2.0
+short_description: Fast Image Edit to Video (IE2V) pipeline.
+---
+
+# 🎬 IE2V Fast Studio: Edit-Then-Animate Pipeline
+
+Welcome to the **IE2V (Image Edit to Video) Fast Studio**. This space bridges the gap between static image manipulation and next-gen video synthesis by combining **FireRed-Image-Edit-1.1** and **Wan 2.2 14B (FP8 Dynamic Quantized & AoT Compiled)** into a unified, high-performance creative sequence.
+
+Modify specific regions or styles of an image with precise natural language instructions, then immediately generate seamless, fluid cinematic motion from the modified canvas.
+
+---
+
+## 🚀 Pipeline Core Architecture
+
+### 🛠️ Phase 1: Local Image Canvas Editing (FireRed 1.1)
+- **Granular Control:** Powered by Qwen-Image-Edit-Rapid, enabling complex workflows like object replacement, style changes, or clothing modification (*e.g., "Convert it to a dotted cartoon style"*).
+- **Flash Attention 3 (FA3):** Specialized hardware acceleration for instantaneous multi-reference canvas modifications.
+
+### 🎥 Phase 2: Ultra-Fast Motion Generation (Wan 2.2 14B)
+- **Lightning LoRA Integration:** Generates striking, stable video outputs in just **4 to 8 inference steps** instead of the traditional 30+.
+- **FP8 Quantization & Graph Compilation:** Uses `torchao` and `spaces.aoti_load` to minimize VRAM latency, maximizing safety boundaries within ZeroGPU infrastructure.
+
+---
+
+## 🛠️ Infrastructure & Resource Management
+
+To avoid memory fragmentation and `RuntimeError: CUDA out of memory` when switching between models, the application strictly isolates the execution contexts.
+
+### VRAM Recycler Pattern (`app.py`)
+Memory is forcefully claimed and flushed back to the OS before and after each inference step:
+---
+title: IE2V Fast Studio (FireRed 1.1 + Wan 2.2 14B)
+emoji: 🎬
+colorFrom: red
+colorTo: purple
+sdk: gradio
 sdk_version: 6.17.3
 app_file: app.py
 pinned: false
 license: apache-2.0
 short_description: Fast Image Edit to Video (IE2V) pipeline.
+tags:
+  - track:backyard
+  - achievement:offgrid
+  - achievement:offbrand
+  - achievement:fieldnotes
 ---
 
 # 🎬 IE2V Fast Studio: Edit-Then-Animate Pipeline
@@ -57,4 +101,24 @@ Wan-AI Team for the foundational Wan 2.2 model weights.
 
 FireRedTeam for the state-of-the-art Qwen-based editing backbone.
 
+cbensimon for the pre-compiled WanTransformer3D AoT modules.
+```python
+def clear_vram():
+    import gc
+    import torch
+    gc.collect()
+    torch.cuda.empty_cache()
+
+Installation Blueprint
+Ensure your runtime environment satisfies the compiled requirements:
+
+pre-requirements.txt: Enforces pip>=23.0.0 for safe wheel building.
+
+packages.txt: Must include system-level ffmpeg for video multiplexing.
+
+🤝 Credits & Acknowledgments
+Wan-AI Team for the foundational Wan 2.2 model weights.
+
+FireRedTeam for the state-of-the-art Qwen-based editing backbone.
+
 cbensimon for the pre-compiled WanTransformer3D AoT modules.