ProCreate, Don't Reproduce! Propulsive Energy Diffusion for Creative Generation

TL;DR

ProCreate is a novel method that enhances diversity and creativity in diffusion-based image generation by actively pushing generated images away from reference images, also preventing training data reproduction.

Contribution

ProCreate introduces a simple, effective technique for increasing diversity and preventing data reproduction in diffusion models, validated on a new few-shot creative generation dataset.

Findings

ProCreate achieves higher sample diversity and fidelity.

It effectively prevents training data reproduction.

Validated on FSCG-8 dataset with diverse categories.

Abstract

In this paper, we propose ProCreate, a simple and easy-to-implement method to improve sample diversity and creativity of diffusion-based image generative models and to prevent training data reproduction. ProCreate operates on a set of reference images and actively propels the generated image embedding away from the reference embeddings during the generation process. We propose FSCG-8 (Few-Shot Creative Generation 8), a few-shot creative generation dataset on eight different categories -- encompassing different concepts, styles, and settings -- in which ProCreate achieves the highest sample diversity and fidelity. Furthermore, we show that ProCreate is effective at preventing replicating training data in a large-scale evaluation using training text prompts. Code and FSCG-8 are available at https://github.com/Agentic-Learning-AI-Lab/procreate-diffusion-public. The project page is available at https://procreate-diffusion.github.io.