Project - A.I. Memories (Ongoing)
IDEA:
I began experimenting with AI image generation, but I wanted to find a ‘use’ for it. To use it as a ‘tool’, to give it a purpose other than just making images for the sake of it. Eventually, I came up with the idea of attempting to recreate images of places and events that I can no longer revisit. The places either no longer exist, or when the events happened, there was no one there with a camera to ‘document’ those events or places.
I wanted to create what appeared to be a photograph of myself standing in the first derelict building that I visited when I was about 17 years old. This building no longer exists, the site has long since been cleared and now a Costa Coffee stands where this house once did. My experience of being here is one that’s been pretty important to me for a long time.
PROCESS / EXPERIENCE:
I must admit to having great difficulty with this idea though, largely because I’ve always felt like I’m ‘fighting’ with the AI image generator all the time, and the process becomes tiring and frustrating. Here however, I continued to persevere. Could I find a way to create an image of a derelict house with a dog basket under the staircase? No! No matter how I phrased, ‘dog basket' or ‘basket’ or whatever other word I tried, the AI image generation software insisted that a dog must exist because after all, it’s a dog basket! If I tried other versions of basket, the AI would add fruit or something else, so I gave up on that. It also insisted that every single feature of the building must appear ‘American’, so another fight involved trying to tell the AI that this house is supposed to represent something from England.
After creating the image, I processed it in black and white. I felt the intention behind this was to visually create an image that closely represented something of myself, my ‘style’ if you will. Once I felt happy with that, I felt that something was missing, as if it wasn’t enough on it’s own. Interestingly, I wonder if the feeling behind this was due to the fact that it’s not ‘real’, I wasn’t really here and I didn’t really take this photograph. It’s not even a photograph after all. That whole experience is completely missing, only the memory of being somewhere similiar to this exists.
I eventually came up with the idea of using an AI voice to narrate the ‘story’ behind the image. I really liked this idea. This allowed me to chose to voice. I quite like the sort of monotone voice here but it’s difficult for me to describe why.
THERAPEUTIC OUTCOME:
The interesting thing is, even though I have my memories of this place, it feels as if this whole experience now ‘lives’ inside the frame of this video. Rather than existing as a collection of still and moving images and sounds in my mind, now they are all here in this little box. In this sense then, everything feels not only contained, but collected, archived, and even tidied up.
I find this beneficial as this helps me to feel like the experience has been effectively organised, and in doing so, ‘dealt with’. I much prefer that these experiences live in these constructed frames. The alternative would be to talk about these experiences, and I can guarantee I can absolutely do that, and gladly for a long time, but it never really feels as if this helps to reach a place of resolution. I’ve certainly spoke about some very difficult things in therapy, but words as a language are not enough for me to provide closure or peace. I find the visual speaks much louder to me. This isn’t to disregard the importance of words, but rather to suggest that the visual speaks a different language; a more subjective one and less restrictive one.
First Video:
The AI images used here were created with various experiments through Bing’s AI image generation. This was my first experiment before moving onto using MidJourney AI image generation.
Further Exploration - Black and White:
Memory: Walking at night to deliver a hand written letter.
The difficulty here was getting the environment right. For me, the environment, the scene, the place, is one of the most important aspects of the memory. The place is where the memory exists for me. Without prompting the AI seems to always assume an American street, so I had to find a way of telling it that this is in Yorkshire. Trying to get the subject to be looking or reading something like a piece of paper was difficult too, as it tended to assume that it was a smart phone, especially as I’d defined the scene as being at night.
Memory: Being shouted at by the teacher in High School
Surprisingly this wasn’t too difficult. One of the important aspect for me was getting the main subject to be pointing directly at the viewer. The AI seems to struggle with hands and hand gestures, so I found myself presented with a lot of random variations. Another difficulty involved the rest of the pupils in the classroom. The more I experimented with various words to describe the emotion of the main subject, the more warped the whole image became until it was so far removed from my original intention. Interestingly too, the AI generated the majority of the pupils as female. Since I didn’t attend an all girls school, this wasn’t what I wanted to represent.
Memory: Birthday party at a friends house
The difficulty here was telling the AI where the camera was placed and how far away from the main subjects it was. Defining the two subjects was difficult too, as the AI seemed to assume that two people, male and female, must be in a relationship. This resulted in the AI being dedicated to placing the two subjects too close to each other, either embracing, kissing, or even in some variations, half naked. All of which just became irritating and annoying. Strangely too, it also insisted that the males were taller than the females. In the end this felt as close as I could possibly manage and in fairness, it’s pretty close to the memory that I have.
Memory: Bored at home.
Getting the expression was important here, but defining it was difficult. After continual attempts I finally stumbled upon a variation that seemed to work. Unfortunately I’ve forgotten the word that I used to explore this. Another difficulty was the environment here. I wanted the subject to feel lost standing in a large living room. In reality the living room wasn’t at all this big, but then I realised that this is how I remember it from the perspective of a child. Although this isn’t a perfect representation of my memory, I feel that the overall mood, lighting, and certain elements all work together well enough for me to consider this, ‘good enough’.
Memory: Getting angry playing with Lego
Memory: Watching ballet
This was incredibly difficult for me. I wanted the boy to be looking through a door with a glass window which seemed impossible for the AI to understand. In the end I stumbled upon this variation which required a lot of extra work in a bid to tweak certain elements of the final image. This was just frustrating.
Colour:
Memory: The Last Day
The difficulty here was specifically in getting the subjects to be stood behind a door with a window that had an autumn leaf pattern etched all over it. I had to use an example of this autumn leaf pattern to give the AI an idea of what I was referring to. Still, it struggled. This feels more like an interpretation of my memory. In that sense I would say this is less successful. In another sense though, the colours of the scene and the lighting work really well to get the ‘feeling’ of the memory right. So it’s a mixed result.
Memory: Watching fireworks.
This was pretty easy to make and I actually enjoyed exploring the variations of the houses and the fences in the background. This was the best variation that was close to my memory. The final result feels a little bit too ‘stock’ though, but who cares? It’s a pretty close representation of memory and so enough.
