Google's Magenta project provides pre-trained models for real-time style transfer using arbitrary images and styles. Check out the full repository here.
In addition to the model, I set up a specific directory structure to match my images. This folder is called Selfies
and is at the root of my google drive. The structure is as follows:
Selfies/
output/
# Output files will be stored here
styles/
style-0.jpg
style-1.jpg
style-2.jpg
style-3.jpg
style-4.jpg
style-5.jpg
style-6.jpg
style-7.jpg
videos/
video-1.mp4
pictures/
img-1.jpg
img-2.jpg
...
The photos in the pictures
directory can have any number of images and any filename as long as the extension is .jpg
.
The code in this colab notebook evaluates 8 styles on the video video-1.mp4
.
Check out the Colab Notebook Here.