The introduction of latent diffusers has brought about a revolution in generative AI and has served as a source of inspiration for creative art. By denoising the latent, the predicted original image progressively brings to life the formation process. However, the diffuser’s denoising nature imposes limitations on the animation, resulting in only a sharpening effect. This research introduces Latent Painter, a technique that utilizes the latent as a canvas and the diffuser predictions as a plan to generate painting animations. Additionally, Latent Painter allows for smooth transitions between generated images, even when using different sets of checkpoints.
The title “Latent Painter” could be rewritten as “The Latent Painting Algorithm” (arXiv:2308.16490v1 [cs.CV])
