WanGPWanGP

Unlocking Local AI Video Generation: A Comprehensive Guide to WanGP

By Technical Expert

The landscape of AI video generation is evolving rapidly. While cloud-based solutions have dominated the headlines, the community has been hard at work making these powerful models accessible on consumer hardware.

Today, we are exploring WanGP (Wan2GP), a specialized interface designed to bring the capabilities of models like Wan 2.1, Hunyuan, and LTX to local machines. This guide aims to provide a friendly, objective walkthrough for setting up and utilizing this tool, specifically tailored for users with varying levels of hardware power—from modest 6GB GPUs to high-end workstations.

Hardware and Software Prerequisites

Before diving into the installation, it is important to ensure your environment is prepared. WanGP is optimized for NVIDIA GPUs, and while it is remarkably efficient, video generation remains a compute-intensive task.

To ensure a smooth experience, please verify the following:

  • GPU Requirements: An NVIDIA RTX 10XX series card or newer is recommended.
  • VRAM Capacity: One of WanGP's strongest features is its efficiency; it can run basic 1.3B parameter models on as little as 6GB of VRAM. However, for larger models (14B), 12GB or more is advisable.
  • Python Version: The codebase is strictly tested on Python 3.10.9. Using other versions may lead to dependency conflicts.
  • Storage & Internet: You will need sufficient disk space for model weights and a stable internet connection to download them upon first launch.

Installation Guide

We recognize that users have different preferences for managing software. WanGP offers two primary pathways for installation.

Option 1: The One-Click Solution (Recommended for Beginners)

For those who prefer to avoid command-line interfaces, we recommend using the Pinokio App. Pinokio acts as a browser for AI applications and handles the environment setup, dependency management, and updates automatically.

  • Download: Visit pinokio.computer
  • Install: Search for WanGP within the app and click install.

Option 2: Manual Installation (For Developers & Power Users)

If you prefer granular control over your environment, the manual installation via Conda is straightforward. This ensures you have a clean, isolated environment for the tool.

  1. Clone the Repository: First, download the source code from GitHub.

    git clone https://github.com/deepbeepmeep/Wan2GP.git
cd Wan2GP

  2. Create the Environment: Set up a specific Conda environment with Python 3.10.9.

    conda create -n wan2gp python=3.10.9
conda activate wan2gp

  3. Install Dependencies: We recommend installing PyTorch first to ensure CUDA support is correctly configured, followed by the project requirements.

    pip install torch==2.6.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu124
pip install -r requirements.txt

For more nuanced troubleshooting regarding installation, please refer to the INSTALLATION.md file included in the repository.

Launching the Interface

Once installed, WanGP operates via a web-based user interface (Gradio).

Basic Launch

To start the standard interface, run:

python wgp.py

Upon execution, you will see a local URL (usually http://localhost:7860). Opening this in your web browser will reveal the dashboard where you can select your desired model from a dropdown menu.

Advanced Launch Modes

WanGP supports command-line arguments to pre-load specific configurations:

  • Image-to-Video Mode: If your primary goal is animating static images. 
    python wgp.py --i2v
  • Low-VRAM Optimization: To force the loading of the smaller, faster 1.3B model immediately. 
    python wgp.py --t2v-1-3B

Understanding the User Interface

The WanGP interface is designed to be intuitive, divided into three logical sections:

  1. Main Generation Panel: This is your command center. It contains the Model Selection dropdown, the Prompt text area (where you describe your video), and the Generate button.
  2. Advanced Settings: By checking the appropriate box, you reveal controls for Steps, Guidance, and Seeds. This allows for fine-tuning the generation process.
  3. Output Gallery: Your generated videos will appear here, ready for playback or download.

There are also specific tabs for LoRAs (for style transfer) and Sliding Window (a technique used for generating longer videos by stitching context together).

Your First Video Generation

Let’s walk through creating your first video. We will use a safe, low-resource configuration to ensure everything is working correctly.

  1. Start the Tool: Run python wgp.py and open your browser.
  2. Select Model: Choose Wan 2.1 text2video 1.3B. This model is highly efficient and runs well on almost all supported hardware.
  3. Configure Settings:
    • Frames: Set to 49 (this results in a video of approximately 2 seconds).
    • Steps: Set to 20. This offers a respectable balance between render speed and visual fidelity.
  4. Enter Prompt: Type: "A cat walking in a garden".
  5. Generate: Click the button and observe the terminal for progress bars.

Once complete, the video will be displayed in the output section.

Deep Dive: Choosing the Right Model

WanGP supports a variety of models, each with distinct strengths and resource demands.

Text-to-Video (T2V)

  • Wan 2.1 T2V 1.3B: The entry-level champion. It requires only 6GB VRAM and generates video quickly. Perfect for testing prompts.
  • Wan 2.1 T2V 14B: A significantly larger model offering superior detail, lighting, and physics. Requires 12GB+ VRAM.
  • Hunyuan Video & LTX Video: Alternative architectures supported by the tool. Hunyuan is noted for high quality but slower generation speeds; LTX is often preferred for longer sequence generation.

Image-to-Video (I2V)

  • Wan Fun InP 1.3B / 14B: specialized in taking a static image and animating it based on a text prompt.
  • VACE: offers more advanced control mechanisms for video generation.

Hardware Recommendations

  • 6-8GB VRAM: Stick to the 1.3B models.
  • 10-12GB VRAM: You can comfortably explore the 14B models or Hunyuan.
  • 16GB+ VRAM: You have the headroom to run any available model and experiment with longer video durations.

Mastering the Parameters

To get the best results, it helps to understand the "levers" you can pull in the Advanced Settings.

Frame Count

This determines the duration of your video.

  • 25 Frames: ~1 second.
  • 49 Frames: ~2 seconds.
  • 73 Frames: ~3 seconds.
  • Note: Increasing frames linearly increases VRAM usage and generation time.

Steps

This defines how many iterations the AI takes to refine the video from noise to a clear image.

  • 15 Steps: Fast preview, but may look "fuzzy."
  • 20 Steps: The standard recommendation for efficiency.
  • 30+ Steps: Higher fidelity, but diminishing returns on time invested.

Guidance Scale

This controls how strictly the AI adheres to your text prompt versus its own creative interpretation.

  • 3-5: Creative, artistic, but may deviate from the prompt.
  • 7-10: Balanced adherence.
  • 12+: Very literal, which can sometimes lead to visual artifacts or rigid motion.

Optimization and Troubleshooting

Even with powerful hardware, users may encounter hurdles. Here are objective solutions to common issues.

Handling "Out of Memory" (OOM)

If your generation fails due to memory limits:

  1. Switch to a 1.3B parameter model.
  2. Reduce the Frame count (e.g., from 73 down to 49).
  3. Ensure Quantization is enabled (this compresses the model to fit in memory).

Improving Generation Speed

If the process feels too slow:

  1. TeaCache: This is a feature that optimizes the caching of video timesteps. Enable it by launching with: python wgp.py --teacache 2.0.
  2. Steps: Lower your steps to 15 or 20.
  3. Sage Attention: Advanced users can install Sage Attention kernels for faster computation (refer to INSTALLATION.md).

Improving Visual Quality

If the output is blurry or incoherent:

  1. Increase Steps to 25-30.
  2. Refine your prompt (see below).
  3. If hardware permits, switch to a 14B model.
  4. Enable Skip Layer Guidance in the advanced settings to fine-tune the model's focus.

The Art of Prompting

The quality of your video output is heavily dependent on your text input. A structured approach yields the best results.

Recommended Structure: [Subject] [Action] [Setting] [Style/Quality Modifiers]

Examples:

  • Cinematic: "A red sports car driving through a mountain road at sunset, cinematic, high quality."
  • Atmospheric: "A cat sitting on a windowsill watching rain, cozy atmosphere, soft lighting."

Expert Tips:

  • Be specific about lighting (e.g., "soft lighting," "sunset," "neon").
  • Define the style (e.g., "realistic," "anime," "oil painting").
  • Use quality keywords like "high quality" or "detailed" to steer the model toward better aesthetics.

Next Steps

We hope this guide has helped you successfully generate your first AI videos with WanGP. As you become more comfortable with the tool, we encourage you to explore more advanced features:

  • LoRA Training: Learn how to fine-tune the model for specific styles or characters (See LORAS.md).
  • VACE ControlNet: Explore advanced video control (See VACE.md).
  • Community: Join the Discord Server to share your creations, view workflows from other users, and get assistance.

Happy creating!

Read More

Original link: Getting Started with WanGP