FLUX on Replicate has been significantly sped up, with optimizations now open-source for transparency and collaboration. The end-to-end speeds for various FLUX models are provided, showcasing impressive improvements in processing time. The collaboration with Black Forest Labs has resulted in faster models, achieved through optimization and the utilization of a new synchronous HTTP API. The quantization method slightly alters model outputs but has minimal impact on quality. By making improvements open-source and collaborating with the AI Compiler Study Group, FLUX aims to provide a fast and efficient solution while encouraging community collaboration. Future advancements in speed are anticipated, promising exciting developments for FLUX users.
https://replicate.com/blog/flux-is-fast-and-open-source