xingd / style2paints

sketch + style = paints !

STYLE2PAINTS

ZhiHu BiliBili

The AI can paint on a sketch accroding to a given specific color style.

Example 1 (Google Search results test)

A content sketch (the first google image search result of key word 'anime sketch') and some style images:

. . .

Results:

. . . .

. . . .

. . . .

Example 2 (western sketch)

A western content sketch and 2 style images:

. . . .

Example 3 (messy sketch)

A messy content sketch and 2 style images:

. . . .

. . .

Example 4 (detailed sketch)

A detailed content sketch and 2 style images:

. . . .

. . .

Example 5 (simple sketch)

A simple content sketch without shadow rendering and 2 style images:

. . . .

. . .

Requirement

pip install tensorflow_gpu
pip install keras
pip install chainer
pip install cupy
pip install bottle
pip install h5py
pip install opencv-python

Launch Server

git clone https://github.com/lllyasviel/style2paints.git
(then download all pretrained models from 'release' page and then put them in 'style2paints/server')
cd style2paints/server
python server.py

Model

Models are avaliable in 'release' page.

1. base_generator.net all rights reserved 2017 style2paints
2. paintschainer.net from paintschainer
3. google_net.net from nico-opendata

Training Datasets

1. The recommended training dataset of illustrations is the 400k images from nico-opendata

2. The recommended training sketches is from sketchKeras

Community

QQ Group ID: 184467946

Paper Reference

The paper is accecped by ACPR 2017.

@article{StyleTansferForAnime,
    Author = {Lvmin Zhang and Yi Ji and Xin Lin},
    Title = {Style Transfer for Anime Sketches with Enhanced Residual U-net and Auxiliary Classifier GAN},
    Journal = {arXiv:1706.03319},
    Year = {2017}
}