kausthubkannan/fungangs-ai repository overview

⁠funGangs.ai

⁠Code

Find code @ repo⁠

⁠Background of the Model

1. The model is trained on StyleGAN and PyTorch is used for the development.
2. The Jupyter Notebook is present in the following Kaggle link⁠.
3. The dataset which is trained upon is Pixelated Treasures: 10K CryptoPunks which can be found via this link⁠.

⁠Features

⁠Image Generation:

The image generation as said uses a StyleGAN custom-trained model.

⁠Story Generation: (TODO)

Currently, the story generation is supported by this Hugging Face API interface⁠. For further addition and gamification, custom trained LLM model will be added.

⁠Database Support:

The future of this project aims to gamify and make it user-specific. Hence, Database support has been added and is in progress to store user-specific data and images.

⁠ToDo

The project further aims to expand to below features:

• FunGangs Image Generation
• Story Generation based on Image-context
• User Interactive Pix2Pix GAN Image Generation
• Wiki API for the Gangs character

⁠API Docs

Currently, the APIs support the generation of images and future expansion is the integration of LLMs for story generations.

⁠Endpoints

Image Generation: /generated/{number_of_images}
The endpoint requires number_of_images to be entered. The response would be a single image which is a collage of a given number of images as shown in the image.

Example:

1. Fungangs Image Generation:

JavaScript

import axios from "axios"

try{
  const number_of_images = 16
  const url = "http://127.0.0.1.8080/generated/{number_of_images}"
  const generated_image = axios.get(url)
}catch(err){
  console.log(err)
}

Python

import requests

try:
    number_of_images = 16
    url = f"http://127.0.0.1.8080/generated/{number_of_images}"
    response = requests.get(url)
    generated_image = response.content
except for Exception as err:
    print(err)

⁠Note

The model, which is based on StyleGAN, requires high memory in terms of CPU and CUDA. Hence, please restrain from generating a number of images more than the range of (8, 48). The story generation is on BETA as it is based on the existing Hugging Face interface. Soon custom LLM support will be added."