Hello internet! Stas and Imi here.
Two weeks into the SCA, and we’re already learning a helluva lot. Apparently, our cohort is the first to have many of our lessons dedicated to learning (or rather trying to come to terms with) AI.
Although I’ve had little experience with AI tools, I was excited to learn how to harness them properly to make something cool. The rules of the brief were simple: remix a portrait of a new subject based on someone else’s gif. Use only the gif or existing files as your source material, include a two-word headline and reveal something new about the subject. Oh, and make it memorable!
As my subject was Charlotte, I sat down with Rose, the original author, to learn more about her. Over a chicken wrap, Rose explained Charlotte’s interests, her inner world, likes and dislikes. From these insights, I began to paint a more vivid picture of her and how I might like to use AI to represent her.
My previous conversations with Charlotte were limited, so I found this exercise interesting, as whatever Rose chose to reveal or withhold would ultimately impact my final portrait. The task was as much about listening and asking the right questions as it was about AI.
I sketched many ideas but ran into problems when I realised they were trying to say too many things at once. I returned to the drawing board and zeroed in on one thing I wanted to reveal about Charlotte. Ultimately, I decided to communicate her nerdy fascination with famed psychologist Sigmund Freud by capturing her in his thinking repose.
However, using the AI tool almost felt like learning to drive. Vague, wordy prompts that did not yield the desired result were akin to violently hitting the curb while the painstakingly slow rendering time caused similar frustration to forgetting to put the keys into the ignition. I learned the more specific the prompt, the better the outcome.
Another handy trick to constructing a compelling AI-generated image was to work from the background to the foreground in manageable chunks. The tool is surprisingly limited and doesn’t do well when it is given more difficult compositions or any subject matter it deems offensive, such as cigars.
When the time finally (dauntingly) came to present our remixed portraits back to the group, I found that transitioning from gif making to jpeg making was quite a selective process. I had perhaps made my gif from Gary’s workshop too busy with various components, which didn’t add anything profound to my gif’s overall message.
Going into the next workshop, I was going to do everything I could to not overcomplicate and dilute my overall message.
So when it came to transferring Zoe’s wonderful gif into a JPEG and also deliberating on 2 (it had to be 2) words to be added in… it had been difficult not to:
a.) Represent the subject in a bland way
b.) Come up with something altogether cryptic
What transpired was that, without philosophising too much, cuts had to be made. Being intentional with choices and taking a simple route in attaching one single AI component (a doorbell) to Honor’s face meant that, unlike my gif, the message of the JPEG was more concentrated. The doorbell stemmed from Honor’s fear of doorbells and I made the decision to place it on her nose to hint at a clown-like appearance; Honour mentioned performing stand-up comedy on a few occasions.
It was a challenge translating the humour and fluidity that naturally comes with gifs into a still image with words, so I eventually settled for a conversational play on what we associate with doorbells: ‘Dong? Ding.’ My change of approach from making an overcrowded GIF to a simplified JPEG reflected in the pleasant response I received for my work, including ‘nice hierarchy!’
Learning from Gary’s stark visual metaphors in the initial gif workshop and Rob’s urge to be selective in Remixed Portraits, the main takeaway emerged: less can be more. Whilst the GIFs were fun to make, there was something more sophisticated about being selective with the endless possibilities of moving images and AI. It’s more fun to tease…
That being said, reviewing everyone else’s JPEGs was thoroughly enjoyable and hilarious and proved how expansive human and artificial imagination can be. Can AI and human behaviour really truly peacefully coexist? Well, many of us grew frustrated at the fact that generative fill often manages to completely reconfigure the clear vision we originally had in our head. This exercise proved how meticulously we had to articulate our prompts to the generative fill engine to deliver a vaguely satisfactory result.
Learning AI software has become more integrated in this year’s programme as opposed to previous years… so it’s going to be interesting to see how utilising these tools, such as Adobe’s generative fill, plays out. I currently regard AI as a bit of an awkward but nonetheless fun third wheel. Things are only about to get scarier and sillier.
Over and out!
Stas and Imi