{"id":966,"date":"2025-04-10T13:25:59","date_gmt":"2025-04-10T13:25:59","guid":{"rendered":"https:\/\/liipetti.net\/visual\/?page_id=966"},"modified":"2025-04-10T13:47:42","modified_gmt":"2025-04-10T13:47:42","slug":"personally-diffused","status":"publish","type":"page","link":"https:\/\/liipetti.net\/visual\/personally-diffused\/","title":{"rendered":"Personally diffused"},"content":{"rendered":"\n

Personally diffused<\/h2>\n\n\n\n

\"\"<\/a>\"\"<\/a><\/p>\n\n\n\n

From 2018 to 2020, I mainly used GAN models that I coded and trained myself, using my own image material, to generate new works which were based on visual features derived from the images used in training. More recently I have tried the same approach using diffusion models. For training I use very small datasets of visually consistent images, and training times are of the order of one or two days. I have released the software in GitHub as urdiffusion<\/a>.<\/p>\n\n\n\n

Urdiffusion is not for everyone, it is difficult to control and absolutely not for realistic images. After all, I developed it in order to explore visuality from my own point of view, according to my preferences. <\/p>\n\n\n\n