I trained a model on childhood photos to simulate memory recall
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
OpenAI model activity is surfacing — worth tracking for capability changes, ecosystem impact, and availability.
A developer fine-tuned the SDXL model using 60 childhood photographs from a limited family archive to simulate memory recall. Instead of faithfully reconstructing the images, the model generates unstable variations, producing familiar-feeling spaces, faces, and fragments that may not have existed. The developer plans to share more experiments, project files, and tutorials through YouTube, Instagram, Patreon, and Uisato Studio.
I fine-tuned the good-old SDXL on 60 photographs from my childhood, using a limited family archive as the dataset through which to revisit that period of my life. Rather than reconstructing those images faithfully, the model produces unstable variations: spaces, faces and fragments that feel familiar without necessarily having existed.
This speculative study treats generative hallucination as an analogue for recollection: not the retrieval of a preserved image, but the reconstruction of a past from incomplete traces. This resonates with contemporary accounts of episodic memory as a reconstructive rather than reproductive process. The model becomes a kind of externalized mnemonic apparatus, situated somewhere between archive, memory and imagination.
Tools used: Kohya, WarpFusion, TouchDesigner, Premiere, After Effects, Ableton Live, Expressive Osmose, Soma Cosmos.
More experiments, project files, and tutorials, through YouTube, Instagram, Patreon, and Uisato Studio.