Arthur Wesley Dow was an American artist from the turn of the 19th and 20th centuries. In his paintings, he actively worked with expressive, simplified forms and silhouettes.


Among all the artist’s works, I was most impressed by the manufactory and urban scenes. In my view, they look the most interesting and best capture the expressiveness that Arthur Doyle aimed for in his time. Based on this criterion, I selected the images for model training—those depicting urban development.
As my final series of generations, I plan to continue the theme of industrialization using an example from one of Russia’s most densely industrialized cities, Magnitogorsk.
The generative model training process
Before training the model, you need to install all the required libraries. I also connected to the drive where the image archive for analysis was stored. At this stage, key settings are established regarding the images being processed—their resolution and the prompt.


The image archive needed to be processed and uploaded to the working environment so that a model could be trained on it later. To do this, I am setting the exact path to the folder using `
local_dir = «/content/drive/MyDrive/pressd»At this stage, the key prompt style for the trained model is also being formulated.


Next, we proceed to train the model. Here, the input and output folders are configured in more detail—specifying what the model will be trained on and where it will be saved. The max_train_steps parameter is also entered, which determines how accurately and thoroughly the model will be trained.
The final stage following successful completion is loading and testing the model in a live environment and conducting test generations.


In the test generations, I checked how accurately the model captured the artist’s key stylistic elements and whether it conveyed the industrial, oppressive mood I intended to achieve to the appropriate degree. With the set number of steps, the model learned to reproduce the results I needed, so I proceeded to generate the final series.
Image Series
