top of page


Do you possess a work of art that you think could be an original? Or do you want to buy a work of art - and be as sure as possible that you are not being taken in by a forger? Every art lover is confronted with these questions again and again.

But when is a work of art actually genuine?

The simple answer to that is: if it is in the catalog raisonné. But the reality is much more complex. In the past, the top experts for a painter or a style have repeatedly allowed themselves to be deceived, for example by counterfeiters such as Wolfgang Beltracchi. In addition, art historians often do not agree, as the write-ups and write-downs of recent years show  - such as Rembrandt's "Man in a Gold Helmet", Vincent van Gogh's "Still Life with Jug and Pears" and Leonardo da Vinci's "Salvator Mundi".

Other authentication methods are not always helpful either. Provenances can be fake. Chemical analyzes say nothing about the authenticity of a work if they are inconspicuous. The painting may then come from the assumed time – but not by the assumed painter. Or a forger used pigments that were commercially available at the time a painting was supposed to have been created. Even aging processes can be accelerated artificially.

In many cases, therefore, whether a painting is actually genuine can only be assessed using the results of a whole series of test procedures and methods – and even then not with 100% certainty.


Color palette: Artificial intelligence does not take spatial structures such as the thickness of the paint application into account. Nevertheless, it is able to identify features in a work of art - even those that remain hidden from the human eye. 

We are now able to offer you a new method that uses artificial intelligence to assign a painting to a painter with a high degree of probability. The basis for this is an ensemble of neural networks that has been trained with thousands of pictures by a wide variety of painters and correctly recognizes the personal style of the artist with an average accuracy of over 90 percent. The method is also constantly being further developed. As with other procedures however, it is unlikely to ever have a 100% reliable verdict.


In contrast to human judgments and expertise, this method has two decisive advantages.

Firstly, the method is interest-free. The algorithm has no feelings and no preferences for an image, neither liking nor disliking an image under test. Secondly, the process is completely transparent. The accuracy of the algorithm in detecting an artist's originals is disclosed, as is the accuracy in detecting counterfeits and imitations of the artist. You can see an example of this transparency in the case studies.

However, the procedure also has limitations. A large number of digital recordings of an artist must be procured for training the algorithms. For the evaluation, imitations and, at best, even proven forgeries of the artist in question are necessary. In addition, the networks for painters who are not yet in the data set must be trained for several days.

We see the authenticity check with artificial intelligence as a promising method. In addition to other methods, in many cases there can be another and often clear indication as to whether a painting can be assigned to a specific artist - or not.


Our mission is to establish artificial intelligence as a valid and recognized art analysis method, not only for authenticity testing, but also for other issues. This requires understanding – and traceability. For this reason, Art Intelligence GmbH attaches great importance to presenting the process as comprehensively as possible - and to clarifying all results as well as the methodology and procedure openly and transparently.


Man and machine: Artificial intelligence is not (yet) able to replace art historians. However, it can provide an additional and, above all, irrelevant indication, especially in the case of contradictory judgments by human experts. 

bottom of page