Chance plays a central role when working with diffusion models in generative AI — and from day one, I’ve been drawn to exploring exactly that. Using a LoRA trained on my own photographs, taken with my Minox 35GT and Nikkormat FTn cameras, I’ve generated over 6,000 images. Rather than crafting detailed prompts, I relied solely on the LoRA’s trigger word, allowing the AI to hallucinate freely based on that influence alone.
From those thousands of outputs, I selected the images that spoke to me. Unlike in Memories Without Origin, this series involved much more hands-on work. During the upscaling phase, I created up to ten variants of each image, each with subtle deviations or prompted for specific changes. From these, I selected the strongest elements and retouched them into a final piece. In the final pass, I added simulated film grain and made colour corrections where I felt it was needed.
As in previous "work", I see myself primarily as a curator, though in this series, perhaps also as a craftsman.
I gave the randomness free rein; I merely selected and enhanced the moments where I saw something worth keeping.