Building AI Art for The Alleys of Old Europe

European Alleyways

In early 2021 we released our latest collection of AI art, called “Alleys of Old Europe”. 

This is the first installment of the ‘Photorealism Gallery’ (name subject to change 😊 ) – namely AI generated art that isn’t trained exclusively on brush-stroke oil paintings, but instead aims for more realistic textures and representations of life – and potentially more hybrid styles.

The origin of the European Alleyways concept was derived from a few pieces of art we had in our homes from trips and vacations to Europe. Throughout much of Europe, travelers will experience areas of narrow pedestrian streets, multiple story buildings and all of them built many centuries ago. When walking through these streets you never know what kind of adventure or discovery you are about to make when rounding the bend. For many Europeans, these narrow streets and alleys are just a daily part of life, but for those of us in North America in particular, they’re incredibly rare, and quite enjoyable to travel, in contrast to our more car-centric freeway lifestyles.

With the Alleys of Old Europe collection, we wanted to reproduce some of that feeling of adventure and intrigue, and see what kind of works the AI system could dream up.  While the system generates a lot of the images with a photorealistic style, there is also some ‘averaging’ of textures in places which gives it more of a smoothed style. So as a result we find there is an interesting mixture of highly realistic portions, alongside more artistic regions, which make the paintings really unique.

The Process

As with most of our AI generated artwork, we started by gathering up as many high-resolution source materials we could find of alleyways, typically sourced throughout Europe, but with a few from Asia and the Caribbean for good measure. One of the trade-offs we always deal with is specificity of subject matter, versus abundance of training material. Typically the more training material one has, the better quality and better variety of images the AI artist can generate. Whether that means a wider gamut of acceptable colors, reflections, times of day, lighting conditions, positioning of foliage and signage etc. – the more examples we can provide the better the outputs from the AI artist.

However if we’re really focused on a particular subject matter, we don’t want the AI artist to learn to produce inconsistent content. So in general, we stuck to the vast majority of training content from European sources, with a sprinkling of 5-10% other regions to help bolster our training data without overwhelming the content and styles too much.

Naming the Works

With the European Alleyways collection, we took a different approach to naming each AI artwork.  We scoured the web for lists of every city, town, village and hamlet throughout Europe as our list of “real city names”.  Then we built another small Generative Adversarial Network (GAN) which would train itself to generate increasingly more realistic names which resemble true names of places in Europe.  So you get names of cool sounding cities like Herbergen and Romstadt which to the best of our knowledge, don't actually exist. So none of the output names are real cities, but they look and sound vaguely similar to real places. We thought this further added to the uniqueness of each piece. Not only is the artistic scene completely unique and never to be produced again, even the name of the town will belong to you, and you alone!

So consider taking a look at the Alleys of Old Europe collection today.


Leave a comment

Please note, comments must be approved before they are published

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.