Reproduce “Lossless Latent Diffusion Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tongda Xu, Yu Zhang, Jun Hu, Huijuan Huang

LLDM addresses the conflict between lossy latent compression, which eases diffusion but loses detail, and pixel-space diffusion. It appends the discarded residual to a VAE latent, scales it down before diffusion, and reverses the scale during decoding for exact reconstruction without changing the diffusion model or objective. On ImageNet 256×256 class-to-image generation, it reports gFID 11.64 without classifier-free guidance; exactness concerns inputs, not generated samples.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.