Co-Creating with AI

Exploring and Integrating Stable Diffusion Models in the Creative Process of Poster Design.


02    Fine-tuning



02 .01     Posters
To fine-tune the Stable Diffusion v1.5 model, I selected a dataset of about 130 images, mainly posters and some flyers. These selections are from the last 30 years, with a small selection from the 50s to the 70s, and they include some of my own designs.
Dataset
↑ Carlo Vivarelli, 1956
↑ Richard Paul Lohse, 1960
↑ Hans Hartmann, 1963
↑ Martin Woodtli, 2000
↑ Ralph Schraivogel, 2006
↑ Wiesmann Daniel, 2009
↑ Albers Inga, 2011
↑ Peter Bankov 2013
↑ Peter Bankov 2013
↑ Wiesmann Daniel, 2016
↑ Roueche Denis, Simon-Vermot Prune, 2016
↑ Geißler Severin, Hofmann Jana, 2017
↑ A. Moritz, A. Massimiliano, J. Bruno, 2017
↑ Personal design, 2018
↑ Franchetti Alice, Cachin Giliane, 2018
↑ Claudiabasel, 2018
↑ Personal design, 2022
↑ Daniel Wiesmann, Robert Radziejewski, 2022
↑ Claudiabasel, 2022
↑ Personal design, 2023

02 .02    Dataset
Interesting areas of these images were selected and cropped with the help of Birme’s website into 512 x 512 px images for fine-tuning and creating various models.

These new images, about 230, were categorised by complexity, from simple to most complex design. The fine-tuning was done with the Dreambooth tool on Google Colab.

Each model was trained with one of these 11 image categories to capture a unique visual style. The following images are a selection from each group.
Dataset, Group 1, for training Model 1

Dataset, Group 2, for training Model 2

Dataset, Group 3, for training Model 3

Dataset, Group 4, for training Model 4

Dataset, Group 5, for training Model 5

Dataset, Group 6, for training Model 6

Dataset, Group 7, for training Model 7

Dataset, Group 8, for training Model 8

Dataset, Group 9, for training Model 9

© 2024 BASEL