Developed and integrated the Representation Autoencoder (RAE) model into the huggingface/diffusers repository, expanding the library’s image encoding and decoding capabilities. The work introduced support for three encoder architectures—DINOv2, SigLIP2, and MAE—paired with a trainable decoder for image reconstruction tasks. Enhancements included robust training scripts, improved configuration handling, and comprehensive documentation updates to streamline onboarding and reproducibility. Leveraging deep learning and image processing expertise with PyTorch and Python, the implementation addressed model instantiation flow and configuration flexibility, while also refining code quality and collaborative workflows to ensure reliability and ease of use for contributors and end users.