Rebecca Cunningham takes a stroll through latent space [Stable Diffusion/ML generated]

This is a surreal slerp through latent space on an LDM trained via Dreambooth on frames from the TV animated series TaleSpin, which some of you might remember from the Disney Afternoon of the early 90s. Some amusing transitions are featured. The interpolation steps are fine enough and the model accurate enough that we even get some nice 'animations' of talking, facial or hand gestures, etc. along the way.

The guided conditioning for each frame is simply the latent embedding for the Rebecca Cunningham character model, while the unguided conditioning (noise) is what is interpolated to change/animate the image. Even though the guided factor is the same for each and every frame, the noise component changing produces a variety of settings, actions, and scenes.

Since these images are synthesized at 768 x 512 pixels, while the model was trained at 512 x 512, that means we get an average of 1½ Rebecca Cunninghams per frame. Huzzah!