Generative Workflow Interview - WhiteRabbitGeometry
Ben Ortlip, known under the artistic Avatar “WhiteRabbitGeometry”, has lived many different lifes, having served in the military, worked in insurance, and creative pursuits such as making art, and writing and performing comedy and music. However, he has had a longstanding interest in AI technology, starting with chatbots in the 1990s, which has led him more deeply to explore the potentials and capabilities of openAI's Dall-e while he was recovering from COVID-19 earlier this year.
Fast forward to today, Ben now uses both Stable Diffusion applications and MidJourney to systematically create images for work on one of his many ongoing creative projects. His inspiration comes from various sources, including personal writings, song lyrics, mathematical equations, and astronomy. In all his creations, it is clear how deeply engrossed he is with the meaning of words and phrases, and how these will be interpreted by the algorithm that he is feeding them to.
For the use of these AI tools, Ben has also developed a highly systemised approach for his workflow process. For example, he might run each line of a poem or song as a prompt and collect the results in thoroughly organised spreadsheets. For using MidJourney, Ben has created a personal Discord server where he utilises the threads function to separate each ongoing project - sometimes up to 40 at a time. The system is both impressive and fanatical, showcasing Ben's dedication and passion for AI image creation.
Check out his work here!
Inspiration is found everywhere for Ben - it may be concepts, words or phrases, that he will use as prompts without much, if any, additional input. These ideas can come from anywhere, but often he finds inspiration in the vast amount of his unpublished writing, especially being fascinated by turns of phrases and the combination or mashup of two or more concepts he finds intriguing.
One of the keys to Ben's workflow and prompt engineering is his interest and understanding of how language is interpreted by machines. He uses his previously gained knowledge and know-how to structure his prompts in a way that allows the machine to better understand his intentions. This means that he will start with the lowest amount of input, only adapting the result once he has seen how the AI has interpreted his initial prompts.
To visualise this process, Ben walked me through an example of an exploratory workflow, which resulted in the Nebula images seen here. The first prompt was purely “Nebula”, which remained in the same spot throughout. In the next step, he first added “Vantablack” and “Barium Sulfide”, which are the purest black and white respectively. This is to guide the output to present the widest contrast value range possible. “Lightpainting”, which is a photographic technique making use of long exposure shots in combination with moving light sources, is also added at this stage, stacked in combination with the aforementioned colours acts to make the light spots “pop”. After this, he decided to also add “Mandelbrot set”, a mathematical term representing a set of complex numbers, which has gained popularity beyond the field for the aesthetic values of the fractal curve’s visualisation.
His process further mirrors his deeply systematic and structural approach and is different to many that I have encountered - he upscales every single image, and only rarely will generate variations or rerolls to achieve the desired result.
It should be of no surprise, that Ben has also built an ethical framework that guides his prompt engineering, of what he calls “ethical prompting”. While he developed the guidelines out of the pursuit for originality, and to support his personal beliefs around the issues of artistic copyright, it nevertheless can act as a counter to the ongoing debate around the use of AI models trained on existing artwork by well-known artists. Ideally offering a more ethical way to reproduce famous styles without directly pointing the algorithm towards a specific person's body of work.
While he himself usually avoids calling for specific art styles or genres overall, relying mainly on the meaning conveyed in the chosen text, he calls for other creators to use descriptors or utilising artistic movements and styles. This, he says, will still enable prompters to achieve the desired stylistic results without the possibility of artistic plagiarism . As the examples show, the style genres can be combined in clever ways making use of semantic hints, similarly to how other prompters often combine multiple artists to achieve new and unique outcomes.
Written by Tina Tiresome
with some help from ChatGTP