End-to-End Training for Unified Tokenization and Latent Denoising

TL;DR

This paper introduces UNITE, a unified autoencoder architecture that jointly trains tokenization and latent diffusion in a single stage, simplifying the process and achieving high-quality results across image and molecule domains.

Contribution

UNITE enables joint training of tokenization and diffusion in one stage, eliminating the need for complex multi-stage training and pretrained encoders, and achieves near state-of-the-art performance.

Findings

Achieves FID 2.12 and 1.73 on ImageNet 256x256 for Base and Large models.

Demonstrates the feasibility of single-stage joint training for tokenization and generation.

Shows that shared parameters foster a common latent language across modalities.

Abstract

Latent diffusion models (LDMs) enable high-fidelity synthesis by operating in learned latent spaces. However, training state-of-the-art LDMs requires complex staging: a tokenizer must be trained first, before the diffusion model can be trained in the frozen latent space. We propose UNITE - an autoencoder architecture for unified tokenization and latent diffusion. UNITE consists of a Generative Encoder that serves as both image tokenizer and latent generator via weight sharing. Our key insight is that tokenization and generation can be viewed as the same latent inference problem under different conditioning regimes: tokenization infers latents from fully observed images, whereas generation infers them from noise together with text or class conditioning. Motivated by this, we introduce a single-stage training procedure that jointly optimizes both tasks via two forward passes through the same Generative Encoder. The shared parameters enable gradients to jointly shape the latent space, encouraging a "common latent language". Across image and molecule modalities, UNITE achieves near state of the art performance without adversarial losses or pretrained encoders (e.g., DINO), reaching FID 2.12 and 1.73 for Base and Large models on ImageNet 256 x 256. We further analyze the Generative Encoder through the lenses of representation alignment and compression. These results show that single stage joint training of tokenization & generation from scratch is feasible.