arxiv:2610.12843

APEX: Asynchronous Prefix Denoising for Extreme Compression

Published on Oct 9

· Submitted by

David D on Oct 14

LTX.io

Upvote

Authors:

Abstract

Fewer latent tokens promise cheaper diffusion. Yet aggressive latent compression creates a reconstruction-generation dilemma: adding channels improves reconstruction but makes generation harder. We introduce APEX, short for Asynchronous Prefix denoising for EXtreme compression, which pairs a hierarchical latent space with asynchronous denoising over channel groups. Our autoencoder orders channel prefixes from structure to detail, while one shared denoiser generates them at staggered local times without adding tokens. On ImageNet-512, APEX outperforms existing methods at every evaluated channel count under 64x and 128x compression. At matched training and guided sampling compute, it also outperforms a 256-token model at every evaluated budget, with larger gains at lower compute. Code is available at https://daviddinkevich.github.io/APEX.

View arXiv page View PDF Project page GitHub 8 Add to collection

Community

daviddink

Paper submitter about 18 hours ago

A single token can have multiple diffusion timesteps at once.

APEX combines a hierarchical latent space with an asynchronous denoising schedule. The model tackles generation in a simpler space first, then adds finer details later.

The result is surprising: at matched training and inference compute, our 64-token model outperforms a 256-token baseline. Using 4× fewer tokens is not just cheaper. With APEX, it produces better images.

APEX also makes channel scaling practical under extreme compression, substantially reducing the generation-quality degradation seen in the other methods we tested.