---
license: mit
---
# LocalVideoDub (LVD Studio) - Linux AI Video Dubbing Utility

An isolated, self-contained Python application where short video clips or photos can be uploaded, and TTS, GFPGAN, and Wav2Lip work together so a user can change the dialog and have the lips match perfectly.

![Application Screenshot](LVDStudio.jpg)

## Features & Highlights
* **Zero System Bloat**: Nothing is installed system-wide. Everything remains self-contained inside the folder.
* **Lightweight Download**: The compressed download is only ~108MB because the required AI models auto-download on the first launch. 
* **Full Setup Size**: The fully installed environment expands to approximately 9GB.
* **Smart Loop**: If your speech text is longer than the original video, the video automatically loops to match the audio length.

---

## Setup and Installation Instructions

> **CRITICAL DRIVE NOTE**: Ensure your installation directory or USB thumb drive is formatted to a native Linux filesystem (like **ext4**). Do **NOT** use spaces in your folder or USB drive names (e.g., use `LVD_Studio`, not `LVD Studio`), or the installer engine will crash.

1. **Download and Extract**
   Download `LocalVideoDub.zip` from the Files tab and extract it.

2. **Open Terminal**
   Open your Linux terminal inside the extracted `LocalVideoDub` project folder.

3. **Build the Isolated Environment**
   Run the installation script to build the local Python 3.10 sandbox and restore all core dependencies:
   ```bash
   ./install_environment.sh
   ```
   *(The first installation can take a while depending on your internet speed.)*

4. **Launch the Application**
   Launch the program with:
   ```bash
   ./launch.sh
   ```
   You will see a lot of terminal text while models and services load. This is normal. On the very first boot, it will automatically download the required AI model weight files (`GFPGANv1.4.pth`, `wav2lip.pth`, and `lipsync_expert.pth`) directly to your local models directory.

5. **Access the WebUI**
   When the server is ready, you will see text mentioning: `Debugger is active`. Open your browser and navigate to:
   ```text
   http://127.0.0.1:5000
   ```
   *(Bookmark it for quick access! To stop the server at any time, press `CTRL+C` in your active terminal window.)*

---

## Basic Usage

1. Load a short video or photo into **"Target Video or Photo"**.
2. *(Optional)* Load external audio if your target video has no speech.
3. Enter your text into **"Replacement Speech Text"**.
4. Click **"Execute Video Generation"**.

The finished video will appear in the WebUI on the left side. Click the video to play it. Generated videos are automatically saved in the `outputs/` folder. You can use the program as many times as you want!

A 720x720 Mona Lisa sample image and matching audio clip are included in the `clips/` folder for immediate testing.

---

## Tips for Better Results

* **Punctuation**: The less punctuation you use, the better the speech generation tends to work.
* **Voice Cloning**: If you use a photo or silent video, you must provide external audio for voice cloning.
* **Video Dimensions**: Best results are usually achieved with videos around 720x720 resolution.
* **Head Movements**: Keep the subject facing forward whenever possible. Minimal head movement produces better lip sync. Side angles and extreme facial rotations will not work properly.
* **Mouth Restrictions**: Avoid videos where the mouth opens extremely wide or moves erratically. Male facial hair can sometimes create texture artifacts around the mouth area.
* **Defaults**: Leave "Seed" and "Acoustic Temperature" at their defaults unless you want to experiment.

---

## Troubleshooting & OOM (Out Of Memory) Errors

If you encounter an OOM error or the system lags:
1. Stop the server with `CTRL+C` in the terminal.
2. Restart the application using `./launch.sh`.
3. Refresh your web browser page.
This usually flushes the GPU VRAM memory and resolves the issue.

### System Tested On:
* **OS**: Xubuntu Linux
* **GPU**: NVIDIA RTX 3050 (8GB VRAM)
* **RAM**: 32GB System RAM

---
*This project is considered stable and feature-complete. Future updates are unlikely unless major issues are discovered. Have fun, and be responsible!*