Use Midjourney to make an art historically inspired piece and be prepared to discuss it in class or on discord next week.

‍

I recently saw a wonderful exhibit showcasing the work of Dominican artist, Firelie Báez. I was particularly captivated, and equally haunted, by her piece Can I Pass?. The powerful gazes of the silhouettes lingered in my mind, and I felt inspired to try and use them as a reference point for this exercise.

‍

I’m also inspired by the work of mixed media and collage artists. Recently, I watched a short film by Ricardo Villavicencio — Do The Evolution and admired his “cut paper” aesthetic. I used a still from his film as a reference for the texture I envisioned for my image. I did a quick sketch as a larger reference of Báez’s piece to have a guide that didn’t include color or fill. I used these four images as references for most of my queries, except when otherwise noted.

I kept the same settings throughout my experimenting. When I tried switching to older model versions, I kept encountering errors. I thought it would be interesting to see what patterns emerged without tweaking the settings too much. The only exception was during the later half of my searches, where I maxed out the "Weirdness."

‍

Based on what I know (very little) about how text-to-image- GEN-AI model’s  are trained, my strategy was to begin with a simple prompt, which ended with a reference to a specific / several famous mixed media artists I like. Then I’d take a few images from this process that I liked, and use those images as the main reference.

‍

‍

Final

‍

Afterthoughts …

It was really interesting to see how much or little the model would reference a photo added, as well as any specific art movement or artist. Even though I had weirdness maxed out, if I included an artists name or for example “surrealism”, it seemed like that tag above all else, was emphasized. Overall, the model struggled to create many results that featured “eyes” without any other facial features, which makes sense to me, as this is a pretty abstract request given that I’m also asking for “heads” and “faces” at the same time, which understandably could be viewed as conflicting based on training data.