All Projects

Art Styleify.

Restyle a photo as Van Gogh, Kandinsky, or three other artists in one click, powered by Google Magenta's pretrained arbitrary-style-transfer model.

Computer VisionArchivedMITMagenta arbitrary-style-transfer5 built-in stylesStreamlit
5
Built-in artists
20
Style reference images
TF Hub
Model source
single forward pass
Inference
Web ApplicationSource Code

Overview

How it works

Google Magenta, not a custom NST pipeline

Style transfer runs through Google Magenta's "Arbitrary Image Stylization" model, loaded straight from TensorFlow Hub and called on the content and style images directly — a fast pretrained network doing one forward pass, not a from-scratch VGG optimization loop trained per image.

Styles

Five artists, twenty reference images

Built-in style options: Van Gogh, Josef Sima, Man Ray, Max Ernst, and Wassily Kandinsky — four reference works each, sourced from WikiArt — plus the option to upload any custom style or content image instead.

Van Gogh
The Starry Night · Wheatfield with Crows · Irises · ...
Josef Sima
Les Ombres · Mer · Storm at Fecamp · ...
Man Ray
Max Ernst
Wassily Kandinsky

Method

A pretrained style-transfer network takes a content image and a style image and returns a stylized result in one forward pass — no per-image training or optimization loop.

  1. 01
    Pick style + content

    5 artists or upload your own

  2. 02
    Magenta model

    single forward pass, TF Hub

  3. 03
    Download result

    stylized image

Stack

TensorFlow HubMagenta arbitrary-image-stylization-v1-256TensorFlowimage tensor pipelineStreamlitupload + style picker UIPILimage I/O

Reproduce it

Local
$ git clone https://github.com/baselhusam/art-styleify.git
$ pip install -r requirements.txt
$ streamlit run main.py
Back to all projectsView source →