One Transformer Fits All Distributions in Multi-Modal Diffusion at Scale

TL;DR

UniDiffuser introduces a unified transformer-based diffusion framework capable of modeling and generating multiple modalities and their combinations simultaneously, achieving high-quality results across diverse multi-modal tasks.

Contribution

The paper presents a novel unified diffusion model that handles marginal, conditional, and joint distributions across modalities with minimal modifications to existing diffusion models.

Findings

Achieves state-of-the-art FID and CLIP scores on multi-modal tasks.

Capable of generating high-quality images, text, and their combinations.

Performs comparably to specialized models like Stable Diffusion and DALL-E 2.

Abstract

This paper proposes a unified diffusion framework (dubbed UniDiffuser) to fit all distributions relevant to a set of multi-modal data in one model. Our key insight is -- learning diffusion models for marginal, conditional, and joint distributions can be unified as predicting the noise in the perturbed data, where the perturbation levels (i.e. timesteps) can be different for different modalities. Inspired by the unified view, UniDiffuser learns all distributions simultaneously with a minimal modification to the original diffusion model -- perturbs data in all modalities instead of a single modality, inputs individual timesteps in different modalities, and predicts the noise of all modalities instead of a single modality. UniDiffuser is parameterized by a transformer for diffusion models to handle input types of different modalities. Implemented on large-scale paired image-text data, UniDiffuser is able to perform image, text, text-to-image, image-to-text, and image-text pair generation by setting proper timesteps without additional overhead. In particular, UniDiffuser is able to produce perceptually realistic samples in all tasks and its quantitative results (e.g., the FID and CLIP score) are not only superior to existing general-purpose models but also comparable to the bespoken models (e.g., Stable Diffusion and DALL-E 2) in representative tasks (e.g., text-to-image generation).