deep-waifu-model

This is a part of the DeepWaifu project.

This repository holds the AI model that converts selfies into waifus.

Credit for the model goes to https://github.com/taki0112/UGATIT and https://github.com/t04glovern/UGATIT

The live version of the dapp can be found here.

🗺 Project overview

This project consists of the following repositories:

• https://github.com/chronologic/deep-waifu-ui
• https://github.com/chronologic/deep-waifu-server
• https://github.com/chronologic/deep-waifu-payment-contract
• https://github.com/chronologic/deep-waifu-model (this repository)

👀 Demo video

💽 Installation

Run conda env create -f environment.yml and then conda activate deep_waifu

(to add new dependencies, change environment.yml and then run conda env update -f environment.yml)

🔧 Model setup

The easiest way is to use the pretrained model provided by https://github.com/taki0112/UGATIT here.

There might be an issue with extracting the file, as mentioned here and here. 7-zip on Windows should be able to handle the file.

Once extracted, the model should be placed in the checkpoint directory.

🚀 Running

Run python server.py to start the server

🔥 Usage

The server exposes a POST /selfie2anime endpoint that accepts Form Data with file containing an image and returns a png of a waifu.