RoAR-VAE: Improving Latent Diffusability through Robust Prediction

Researchers propose RoAR-VAE, a visual tokenizer method that aims to improve latent diffusability through robust prediction, according to the paper's abstract. The work revisits REPA-E, a representative approach to improving latent diffusability via semantic alignment or joint tokenizer-diffusion training, and reports a finding about that method. The paper argues an effective visual tokenizer must preserve image fidelity while producing latents that downstream diffusion models can learn efficiently.

An effective visual tokenizer must preserve image fidelity while producing latents that downstream diffusion models can learn efficiently. Prior work improves latent diffusability through semantic alignment or joint tokenizer--diffusion training. Revisiting REPA-E, a representative approach, we find