Perturbations enables the exploration of the latent space around a seed. Perturbing the noise from an initial seed towards the noise from a different seed illustrates the variations in images obtainable from a local region of latent space. Using a small perturbation size produces target images that closely resemble the one from the initial seed. Larger perturbations traverse more distance in latent space towards the second seed, resulting in greater variation in the generated images.

Additional Controls:

Number of Perturbations:

Specify the number of perturbations to create, i.e., the number of seeds to use. More perturbations produce more images.

Perturbation Size:

Controls the perturbation magnitude, ranging from 0 to 1. With a value of 0, all images will match the one from the initial seed. With a value of 1, images will have no connection to the initial seed. A value such as 0.1 is recommended.