dvornikita / PixarAvatar

Generates Pixar-style avatars from your photos

PixarAvatar

PixarAvatar is a project that turns your portrait photos into Pixar characters.

Features

• Given a portrain photo, the application will generate an avatar that resembles the person in the photo.
• Has the ability to analyze the input photos (using a VLM) to generate a more accurate avatar.

Installation

1. Clone the repository:

   git clone https://github.com/ndvornik/PixarAvatar.git

2. Install the required dependencies:

   cd PixarAvatar
   
   # create a conda environment
   conda create -n pixaravatar python=3.9
   conda activate pixaravatar
   
   # install the required packages
   pip install -r requirements.txt

Usage

1. To generate 4 different avatart (to chose from) from a given image, run the following command:

   python generate_avatar.py --image_url=<path_to_image> --output_dir=<output_directory> --num_avatars=4

   Tip: if the generation is slow (takes minutes), make sure you are using a GPU.

2. Parameters to the generate_avatar.py script are as follows:

   • --image_url: The path to the input image.
   • --output_dir: The directory where the generated avatars will be saved. Default is ./output.
   • --num_avatars: The number of avatars to generate. Default is 1.
   • --crop: The way the image will be cropped. Options are [center, top, bottom or none]. Default is center.
   • --generate_prompt_from_image: If True, the the image will be analyzed with a VLM to generate a more specific prompt for avatar generation. Usefull for generating more faithful avatars but slows down the process. Default is False.
   • --device: The device to run the model on. Options are [cpu, cuda]. If not set, the model will run on the device available.

Notebook

The pixar_avatar.ipynb notebook contains the code to generate the avatar and visualize the results.