Prompts Didn’t Make These: Deep Style is Still Here
The first wave of generative AI content was roughly from 2014–2021, before Dall-E and other text-2-image (prompt-based) generators reached the public. Deep style refers to deep learning and neural style transfer that, according to TensorFlow is an “optimization technique used to take two images — a content image and a style reference image (such as an artwork by a famous painter) — and blend them together so the output image looks like the content image, but “painted” in the style of the style reference image.”
The style I applied for my first AI-generated images was a pattern from a West African wax-printed fabric. Vintage 19th century tapestries and lace patterns were also a favorite source for image styles. The source photo was my own. From 2016 to 2019 I used whatever inspired me to create (generate) images, using Deep Dream Generator’s Deep Style tool. Over the years, the results showed more comprehensive knowledge and skill using the tool. It launched a practice that included posting at least one imager every day on social media platforms, which resulted in a increasing number of followers who appreciated the work.
A fan used the above image as an underlay, then they drew over it on an iPad, sent it to an embroidery machine. They were working with the limitations of the embroidery software. The final work was presented to me in the form of a light box (see below). During the same year I was commissioned by a Smithsonian curator to create 11 “featured futurists” for a exhibition. For the image styles, I used images of the subjects’ inventions (see Keller portrait below). This includes antique fabrics and even Braille writing (dot patterns).
At that time, prompt-based text-2-image AI generators were not available. Prompts were not used to create these works. Within the next 6 months, text-to-image models and tools were released to the public resulting in an AI boom. In 2022, the output of state-of-the-art text-to-image models — such as OpenAI’s DALL-E 2, Google Brain’s Imagen, Stability AI’s Stable Diffusion, and MidJourney. However, even with the popularity of the latter tools I didn’t abandon DDG’s Deep Style technique.
Text-to-image models are generally latent diffusion models, which combine a language model, which transforms the input text into a latent representation, and a generative image model, which produces an image conditioned on that representation. The most effective models have generally been trained on massive amounts of image and text data. — wikipedia
In spite of the massive amounts of images and data made publicly available on the Web, arguably without some artists’ consent, there continues to be a dearth of diverse images of people of color. Lack of representation is one of the primary reasons creators from historically marginalized groups remain disengaged from developments in emerging technologies.
In a sense, generative AI and GenAI art is no different from past tech innovations but this is changing due to the proliferation and accessibility of the tools. Also, the image quality and possibilities are improving. The main difference between the above image(s) and the first one I made using Deep Style (in 2016) is the resolution. And it keeps getting better.
Note: The use of Deep Style makes the Text-2-Image output more distinctive but I only use this tool as needed. Sometimes I use Adobe Photoshop to finish AI-generate images.