Every Python developer knows some or all of these libraries, because they’re stable, reliable, and excellent at what they do.

We address a fundamental question: Can latent diffusion models and their VAE tokenizer be trained end-to-end? While training both components jointly with standard diffusion loss is observed to be ...