How to Forge Art

Title of article: “Exploring the structure of a real-time, arbitrary neural artistic stylization network”.
My translation: “Your computer can make fake art”.

For example, if you feed the neural network a picture of a vase (left) and one of Van Gogh’s paintings (middle), it outputs the vase in Van Gogh’s style (right):

(TensorFlow has a tutorial that does this for you.1)

I wanted to see how good the imitations were, just in case I decide to pursue a life of crime as an art forger.

So I chose ten artists (the ones we covered in my 20 days of high school art history).

For each artist, I flipped a2 coin. If the coin came up heads, I found a (not very well-known) painting by this artist. If the coin came up tails, I instead applied the neural network to a stock photo to make it look like the artist’s style.

The resulting images are:

1. Pablo Picasso:

   

2. Caravaggio:

   

3. Leonardo da Vinci:

   

4. Rembrandt:

   

5. Salvador Dalí:

   

6. Georges Seurat:

   

7. Pierre-Auguste Renoir:

   

8. Michelangelo:

   

9. Claude Monet:

   

10. Jackson Pollock:

    

I know which paintings are real and which are forgeries. But I want to know how convincing the fakes are, so I’d like you to take this survey:

Loading…

(There ought to be a link to the answers at the end of the survey, but the survey might crash, so the answers are here.)

Footnotes:

1

Indeed, most of what I know about the algorithm comes from the tutorial. I’ve only skimmed the article, and I don’t understand it yet.

2

digital