FluxSpace: Disentangled Semantic Editing in Rectified Flow Models

Rectified flow models have emerged as a dominant approach in image generation, showcasing impressive capabilities in high-quality image synthesis. However, despite their effectiveness in visual generation, understanding their inner workings remains a significant challenge due to their ``black box'' nature. Recent research has focused on identifying a representation space that facilitates semantic manipulation of generated images, but these models generally lack a GAN-like linear latent space, that allows straightforward control over image generation. In this paper, we introduce FluxSpace, a domain-agnostic image editing method leveraging a representation space with the ability of controlling the semantics of images generated by rectified flow transformers, such as Flux. By leveraging the representations learned by the transformer blocks within the rectified flow models, we propose a set of semantically interpretable representations that enable a wide range of image editing tasks, from fine-grained image editing to artistic creation. This work both offers a scalable and effective image editing approach and significantly enhances the interpretability of rectified flow transformers.