Any-to-Any Generation via Composable Diffusion

TL;DR

This paper introduces Composable Diffusion (CoDi), a versatile generative model that can produce any combination of modalities like text, images, videos, or audio from various inputs, even without specific training data for those combinations.

Contribution

CoDi is the first model to enable flexible, multi-modality generation from arbitrary input combinations by aligning modalities in shared spaces and employing a novel composable diffusion strategy.

Findings

Outperforms or matches state-of-the-art in single-modality synthesis.

Can generate multiple modalities simultaneously and in parallel.

Works effectively even without training data for specific modality combinations.

Abstract

We present Composable Diffusion (CoDi), a novel generative model capable of generating any combination of output modalities, such as language, image, video, or audio, from any combination of input modalities. Unlike existing generative AI systems, CoDi can generate multiple modalities in parallel and its input is not limited to a subset of modalities like text or image. Despite the absence of training datasets for many combinations of modalities, we propose to align modalities in both the input and output space. This allows CoDi to freely condition on any input combination and generate any group of modalities, even if they are not present in the training data. CoDi employs a novel composable generation strategy which involves building a shared multimodal space by bridging alignment in the diffusion process, enabling the synchronized generation of intertwined modalities, such as temporally aligned video and audio. Highly customizable and flexible, CoDi achieves strong joint-modality generation quality, and outperforms or is on par with the unimodal state-of-the-art for single-modality synthesis. The project page with demonstrations and code is at https://codi-gen.github.io