Researchers at Google DeepMind introduced Semantica, an image-conditioned diffusion model capable of generating images based on the semantics of a conditioning image. The paper explores adapting image generative models to different datasets. Instead of finetuning each model, which is impractical for large-scale models, Semantica uses in-context learning. It is trained on web-scale image pairs, where one random image from a webpage is used to condition the generation of another image from the same page, assuming these images share semantic traits
Download the medial app to read full posts, comements and news.