Obsolescent: Reimagining an old photography series with new technology
An exploration of how a photo series captured 10 years ago can be brought to life and transformed in new and novel ways. But what does this new tech mean for the future of photography and art?
In 2014, I took a series of photos across Iceland, shot on 35mm Purple Lomochrome film. It was a collection of about 30 photos, intended to make Iceland even more surreal looking than it already is. Ten years on, and with all the new technology that has emerged, I started to reflect on the place of photography in the new age. The series was part of a group exhibition, ‘Obsolescent’, which was exploring the disuse of film as a medium in the digital age.
Now, in the AI age, what will the impact be on digital photography? Was it, too, becoming obsolescent? I explored this question in an earlier article, through an initial experiment attempting to recreate one of the original photos from the series in an AI generator.
The photo on the left is my original, on the right is an image created entirely through a text prompt (…a ‘word only’ prompt):
Perhaps in reaction to my own article, I wanted to push this further. Instead of this medium becoming obsolescent, how might it be extended or transformed by AI? How could it be pushed further into new realms, and be experienced in novel ways?
I started playing with the AI models, and landed on two main experiments: to bring the photos to life by adding motion to them using generative AI tools, and to create new ‘photos’ in the style of my original photos.
Part 1: Bringing my photos into motion
In part 1 of this exploration, I decided to feed my photos into an AI model, specifically Leonardo.AI. I figured, most of them are on the internet so they’ve probably already been eaten and extracted by generative AI models anyway. May as well experiment with it.
For each photo, I then used generative AI to create a sense of motion. This transformed the series into something entirely new - a dynamic, moving landscape. When I shared the final video with someone, they mentioned it was like “giving those original moments a few seconds”.
Here is the final output:
(…ps if you want to skip my little preamble, you can head straight to the 1:15 mark in the video)
It was kind of wild for me to re-explore this exhibition, and it feels like this process might become a means of world building and transforming photos into whole new mediums.
Side note: I made the sounds for this video using Elevenlabs for added effect.
Part 2: The ‘photos’ I never took
I had a conversation recently with someone who referred to an output from an AI image generator as a ‘photo’. But is it really a ‘photo’? The word photography comes from two Greek words, phos meaning ‘light’ and graphe meaning ‘drawing’. So photography is to draw with light: a process of capturing light as it falls on the objects around us. A moment in time, preserved as a visual history.
AI image generators do not work in this way. Instead, millions of photos and images are used to train an AI model, and then go through some pretty scientific processes with neural nets to be able to create an image from scratch. Text prompts can be used to create a new image that mimics the aesthetic and style of photography and images initially used in the training data.
It's not quite as romantic as drawing with light, is it?
So no, an image generated by an AI is not a photograph, no matter how realistic it appears.
I had a thought recently. If photography is for capturing what is real, are image generators to capture what is not real, and rather, what can be imagined? This led me to Part 2 of my experiment: The ‘photos’ I never took.
The process
There were parts of Iceland I didn’t get to on this trip which I would have liked to have visited.
This inspired the second experiment: to imagine what those places might have looked like, and specifically, whether I could get an AI model to mimic the particular style of the series (i.e. shot in 35mm Purple Lomochrome film)?
I first tried training an AI model on my photos and generating images of famous landmarks using that training data. This was an unsuccessful approach, and it completely missed the style of photography.
Instead, I found better success using text prompts to describe the desired style and landmark. Some of the reference landmarks included the mountain range of Vestahorn, the geyser of Strokkur, the lava fields of Eldhraun, the geothermal pools of the Blue Lagoon, and the black sands of Reynisfjara:
My reaction to this part of the experiment was initially being in awe of how close they get to looking like real ‘photos’. But, they don’t look like my photos, nor do they look like they belong in that particular series (despite the purple and magenta hues). I feel they are too perfect.
This makes sense when considering the vast amounts of photographs on the internet that have been used as training data for the models are from a period where we have used digital technologies to edit our photos and smooth out the aberrations. Camera quality is now incredibly impressive, even just in our phones. In addition, filters are often applied to make adjustments automatically to the photo for us (autocorrecting those ‘terrible’ abnormalities and imperfections out!). AI, in predicting the ‘next best pixel,’ gravitates toward the average photo. If the average photo in the training data is polished and perfect, that’s what it will create.
Since I’ve always leant towards imperfect photos, shot on film, with grain and abnormalities, it is probably harder for an AI model to immediately replicate this particular style (…however, in the future I suspect this will be much easier).
The images generated by the AI model are beautiful. They could easily have been taken by someone else as an edited digital photo. In some ways, it appears that the constant strive towards sharpness and perfection in photography during the digital age has made it incredibly easy to replicate that style.
Obviously, I don’t feel the same connection to the generated images as I do to the originals - they captured moments I lived and experienced. However, would an audience feel the same? Does the viewer, without the connection to the moment, after a certain artistic fit for their wall, perceive the output differently?
And what is the value of art, specifically photography, in an age where image generator outputs so easily mimic an original photo? Perhaps there will be increased value placed on the story of how something comes into existence, or new styles of photography will emerge/be revived that have less focus on perfection. Or maybe we will we go back to more analogue ways?
The problem of AI image generators
We are now awash with AI generated images flooding our feeds, testing our perception of what is real and what is not real. I see videos of people hugging polar bear cubs, riding whales, and so many other ridiculous videos and images. I know they are not real, but there is a moment before I decide it is not real. But what about future generations, and the younger generations (and perhaps even the older generations)? How will they navigate reality?
We were already in Plato’s cave, and now we are sliding even deeper.
Then there is the other ethical aspect. These models have been trained under a veil of ‘fair use’ to be in line with copyright laws. But ultimately, in order to create these models, millions of artists and photographers photos have been used. Does the concept of fair use hold up in this modern age, or is it time to revisit copyright law? What might a ‘fairly trained’ AI model look like in the future?
Final Thoughts:
Do we need a new word for the outputs from image generators?
All of this has me reflecting on whether we need new words to navigate and sense-make in this new era.
The term photograph is clearly not correct. And the word image doesn’t feel like enough. And ‘an AI generated image’ is quite the mouthful. So what words could we explore? Here are some ideas, using the word graphe meaning ‘to draw’.
Mimograph. To draw from mimicry.
Pseudograph. To draw what is not real.
Promptograph: To draw from prompts.
Repligraph. To draw through replication.
Eidograph. To draw from an idea.
Imagograph. To draw from imagination.
Admittedly, I don’t think any of these will catch on. But we do need to start thinking of how we make sense of the new world of technology, and how it will filter down into society.
Until next time.