A state before the picture
Generated images are playing an increasingly important role in our everyday lives – the question “is this AI or is it real?” keeps coming up. But what exactly do generated images actually show if they have no connection to what they show? To answer this question, we must first open ourselves to the idea that generated images are ultimately no longer photography in the traditional sense – or are they?
If we look back a few decades, we realize that photography, as the successor to painting, initially interacted directly with its environment. In analogue photography, a moment materializes in the form of light that reacts with chemistry: Each negative was at the scene, the camera lens saw, light and shadow were preserved. In digital photography, the whole thing is somewhat more abstract, as the medium of film has been replaced by a sensor: Light, colors, objects become signals from which an image is created. It is no longer concretely depicted, but created.
Returning to the generated images, we notice that the location, the moment and ultimately the camera are replaced by a prompt – a text that describes what the AI user wants to visualize. What happens next is in stark contrast to the processes described above: An AI that has been trained with billions of images from archives, platforms and repositories attempts to generate a new image from these image fragments – composed of learned probabilities that could correspond to the prompt. The basis is therefore no longer what is seen, but the recombination of fragments, which is formed into an impression.
If we finally look at the question of what is contained in the training material, we realize that anything that circulates online can appear: Product images, vacation photos, selfies, artwork, x-rays, pornography. Nothing is too trivial or too critical not to become part of a model as a fragment. In the end, the AI generates new images from the noise of past moments as if they had been seen.
The path through this very process is the core of the work, in which the images shown emerge from an intermediate step of image generation. The noise thus consists of fragments of what could later become an image. We see neither narratives nor reality, but traces, surfaces and everything we think we recognize if we look long enough in the pictures. A state before the picture.