Run SkyReels V1 Hunyuan I2V on Windows: Step by Step Guide

SkyReels-V1-Hunyuan-I2V is an advanced open-source video generation model developed by SkyworkAI, designed to facilitate high-quality video production through innovative machine learning techniques.

This model is particularly notable for its capabilities in both text-to-video (T2V) and image-to-video (I2V) generation, making it a versatile tool for creators looking to produce engaging visual content. The underlying architecture of SkyReels is based on fine-tuning the Hunyuan video model, which has been trained on a vast dataset of high-quality film and television clips.

This article will provide a comprehensive guide on how to run SkyReels-V1-Hunyuan-I2V on Windows, covering installation, configuration, and practical usage.

System Requirements

Before diving into the installation process, it is essential to ensure that your system meets the necessary hardware and software requirements:

Hardware Requirements

Graphics Card: A minimum of 16GB VRAM is recommended for optimal performance. NVIDIA RTX 4090 or similar GPUs are ideal.
CPU: A multi-core processor will enhance performance during video generation.
RAM: At least 16GB of RAM is advisable to handle large datasets and processes efficiently.
Storage: Sufficient disk space (at least 50GB) for the model files and generated videos.

Software Requirements

Operating System: Windows 10 or later.
Python: Version 3.10 or higher.
CUDA: Version 12.2 for GPU acceleration.
Pip: Python package installer should be available.

Installation Steps

To successfully run SkyReels-V1-Hunyuan-I2V on Windows, follow these detailed steps:

Step 1: Install Python and Pip

Download and install Python from the official website.
During installation, ensure that you check the option to add Python to your PATH.
Verify the installation by opening Command Prompt and typing:bashpython --version
pip --version

Step 2: Install CUDA

Download the appropriate version of CUDA from NVIDIA's official site.
Follow the installation instructions provided by NVIDIA to set up CUDA on your system.

Step 3: Clone the Repository

Open Command Prompt and execute the following commands to clone the SkyReels repository:

bashgit clone https://github.com/SkyworkAI/SkyReels-V1
cd SkyReels-V1/skyreelsinfer

Step 4: Install Dependencies

Once inside the cloned directory, install the required Python packages using pip:

bashpip install -r requirements.txt

This command will install all necessary libraries for running SkyReels.

Configuration Settings

After installing the necessary components, you may need to configure certain settings before generating videos.

Model Selection

SkyReels offers two main models for video generation:

Text-to-Video (T2V): For generating videos from textual descriptions.
Image-to-Video (I2V): For creating videos based on input images.

Select the appropriate model based on your project needs.

GPU Configuration

If you have multiple GPUs, you can optimize performance by specifying which GPUs to use in your commands. The model supports parallel inference, allowing for faster video generation when multiple GPUs are utilized.

Running Your First Video Generation

To generate your first video using SkyReels-V1-Hunyuan-I2V, follow these steps:

Step 1: Prepare Your Prompt or Image

For T2V tasks, prepare a descriptive prompt that starts with "FPS-24," followed by your desired scene description. For I2V tasks, have an image ready that you want to convert into a video.

Step 2: Execute the Command

Open Command Prompt in the skyreelsinfer directory and run one of the following commands:

For Text-to-Video (T2V)

bashpython3 video_generate.py \ --model_id Skywork/SkyReels-V1-Hunyuan-T2V \ --task_type t2v \ --guidance_scale 6.0 \ --height 544 \ --width 960 \ --num_frames 97 \ --prompt "FPS-24, A cat wearing sunglasses and working as a lifeguard at a pool" \ --embedded_guidance_scale 1.0

For Image-to-Video (I2V)

bashpython3 video_generate.py \ --model_id Skywork/SkyReels-V1-Hunyuan-I2V \ --task_type i2v \ --image path_to_your_image.jpg \ --guidance_scale 6.0 \ --height 544 \ --width 960 \ --num_frames 97 \ --embedded_guidance_scale 1.0

Step 3: Monitor Progress

The command line will display progress logs as the model generates your video. Depending on your hardware configuration, this process may take some time.

Performance Optimization Tips

To maximize efficiency when using SkyReels-V1-Hunyuan-I2V, consider these optimization strategies:

Utilize Multiple GPUs

If available, leverage multiple GPUs for parallel processing. This can significantly reduce video generation times:

Single GPU: Approximately 889 seconds for a short video.
Four GPUs: Reduced to about 293 seconds.
Eight GPUs: Further optimized to approximately 159 seconds.

Adjust Resolution and Frame Count

Experiment with different resolutions and frame counts based on your project requirements and available VRAM. Lowering these values can lead to faster processing times but may affect video quality.

Use Quantization Techniques

Enable FP8 weight-only quantization and parameter-level offloading in your command line options to reduce GPU memory usage without compromising quality.

Troubleshooting Common Issues

While running SkyReels-V1-Hunyuan-I2V, you may encounter various issues. Here are some common problems and their solutions:

Installation Errors

If you face issues during installation:

Ensure that Python and pip are correctly installed.
Check that all required dependencies are listed in requirements.txt.

GPU Not Detected

If your GPU is not recognized:

Verify that CUDA is correctly installed.
Ensure that your graphics drivers are up-to-date.

Slow Generation Times

If video generation is slower than expected:

Check if multiple GPUs are being utilized effectively.
Consider reducing resolution or frame count if VRAM limitations are reached.

Conclusion

Running SkyReels-V1-Hunyuan-I2V on Windows opens up exciting possibilities for creators interested in advanced video generation techniques. By following this comprehensive guide, users can set up their environment efficiently and begin producing high-quality videos using either text or images as input.