AI Art Gallery

Background

Common AI methods for the analysis of images are based on Deep Learning.
For that, neural networks (especially Convolutional Neural Neworks) consisting of several layers are used. The layers themselves, are – vividly considered – filters with different degrees of abstraction: while the “lower” layers rather recognize points, lines, texture patterns etc. – i.e. the “style” – the higher layers serve to recognize the contents of partial or complete images.

This property of neural networks is used in neural style transfer to generate new images. For this purpose one combines an image, which contributes the content, with an image, which contributes the style of the final image.
If, for example, portrait photos are used as content images and paintings of well-known artists as style images, this process can be used to generate images that in part look as if the respective artist had painted the portrait.

Based on these existing concepts we went one step further. Our goal was not to imitate different artists, but to use the described procedures for completely new pictures at the interface between photographies and paintings.
For this purpose, we concentrated mainly on landscape photos in the “content images”, which seemed particularly suitable to us. As “style pictures” we deliberately did not use paintings, but further pictures with mainly scientific-technical background, such as crystals, circuit boards or plants. In our opinion, which has become more and more consolidated in the course of time, each of them has its own “style”, which harmonizes more or less strongly with different landscapes (one can also speak here of “style resonance“) – or not.

The following pictures show a small part of what is possible with the mentioned methods. We are convinced that a completely new art direction will open up here and are happy to be among the first who follow this path.

Art Gallery

Below is a selection of our AI-generated images. All images can be purchased license-free by companies and other organizations in higher resolution. If you have further questions, we are of course happy to help don’t hesitate to write us an email.

Contact

Dr. Dimitrios Geromichalos
Founder / CEO
RiskDataScience UG (haftungsbeschränkt)
Theresienhöhe 28, 80339 München
E-Mail: riskdatascience@web.de
Telefon: +4989244407277, Fax: +4989244407001
Twitter: @riskdatascience