The most capable of the models in the new family, called Flux.1 Kontext, can be prompted with text and, optionally, a reference image to create new images, writes Black Forest Labs in a blog post.