Diffusion Domain Expansion: Learning to Coordinate Pre-trained Diffusion Models

2026-05-22

AI Summary

Existing pre-trained diffusion models struggle to generate content beyond their training scale or under complex conditions. To address this limitation, this work proposes a lightweight, trainable coordination network that fuses the denoising outputs of multiple pre-trained diffusion models, enabling cross-domain and large-scale content generation. The coordination mechanism features a simple yet general architecture, allowing effective out-of-distribution generalization without retraining the underlying foundation models. Experimental results demonstrate that the proposed approach significantly outperforms existing collaborative generation methods in both qualitative and quantitative evaluations on tasks such as long-form audio synthesis and conditional image generation.

Abstract

In this paper, we propose Diffusion Domain Expansion (DDE), a method that efficiently extends pre-trained diffusion models to generate larger objects and handle more complex conditioning beyond their original capabilities. Our method employs a compact trainable network designed to coordinate the denoised outputs of pre-trained diffusion models. We demonstrate that the coordinator can be universally simple while being capable of generalizing to domains larger than those observed during its training time. We evaluate DDE on long audio track generation and conditional image generation, demonstrating its applicability across domains. DDE outperforms other approaches to coordinated generation with diffusion models in qualitative and quantitative evaluations.

Problem

• diffusion models
• domain expansion
• pre-trained models
• conditional generation
• scalability

Innovation

• Diffusion Domain Expansion
• pre-trained diffusion models
• coordinated generation
• domain generalization
• conditional generation