Spaces:
Running
Running
File size: 2,721 Bytes
6275299 38e41fd 6275299 38e41fd 6275299 38e41fd 6275299 38e41fd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 | ---
title: Video Face Replace
emoji: 🎭
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: 6.10.0
app_file: app.py
pinned: true
license: mit
short_description: Swap any face into any video using InsightFace + GFPGAN
---
# 🎭 Face Fusion — AI Video Face Swap
Swap any face into any video using state-of-the-art AI. Built on **InsightFace**, **inswapper_128**, and **GFPGAN**.
> ⚡ This is a **Kaggle notebook project** — run it there for full GPU-accelerated face swapping.
> The Gradio UI is embedded inside the notebook (Cell 4).
---
## ✨ What It Does
Upload a single clear photo of a face, point it at a YouTube video (or upload your own), and the pipeline replaces **every detected face** in the video with your source face — enhanced, sharpened, and re-merged with the original audio.
**Pipeline:**
```
Source Face Photo → InsightFace Embedding
↓
Target Video (YouTube URL or upload)
↓
Per-frame: Face Detection → Swap (inswapper_128) → GFPGAN Enhancement
↓
Audio Merge (ffmpeg) → Final MP4
```
---
## 🚀 How to Run (Kaggle)
1. Open `Face_replace.ipynb` on [Kaggle](https://www.kaggle.com) with **GPU accelerator** (T4 or P100)
2. **Cell 1** — installs packages + downloads models (~1 min)
3. **Cell 2** — set source face path + YouTube URL, downloads & trims video
4. **Cell 3** — processes all frames → outputs `face_swapped_output.mp4`
5. **Cell 4** — launches interactive Gradio UI
> Processes ~12 frames/second on T4 GPU
---
## 🧰 Tech Stack
| Component | Role |
|---|---|
| InsightFace `buffalo_l` | Face detection + recognition |
| `inswapper_128.onnx` | Neural face-swap model |
| GFPGAN v1.4 | Face restoration & enhancement |
| OpenCV | Frame-level video I/O |
| ffmpeg | Video codec + audio merge |
| yt-dlp | YouTube video download |
| Gradio | Interactive UI |
| ONNX Runtime GPU | GPU-accelerated inference |
---
## 💡 Tips for Best Results
- Use a **clear, front-facing photo** — no sunglasses or heavy shadows
- Keep target videos **under 30 seconds** for fast iteration
- Single-face videos give the cleanest results
- Enable GFPGAN (`ENABLE_ENHANCEMENT = True`) for sharper output
---
## ⚠️ Disclaimer
For **creative, educational, and research purposes only**.
Do not use to create deceptive, harmful, or non-consensual content.
Always obtain permission before using someone's likeness.
---
## 🙏 Credits
- [deepinsight/insightface](https://github.com/deepinsight/insightface)
- [TencentARC/GFPGAN](https://github.com/TencentARC/GFPGAN)
- [ezioruan/inswapper_128.onnx](https://huggingface.co/ezioruan/inswapper_128.onnx)
---
*Made with ❤️ by [Aastha Malik](https://github.com/aastha-malik)*
|