Prompt Refinement with Image Pivot for Text-to-Image Generation

TL;DR

This paper introduces PRIP, a novel zero-shot prompt refinement method for text-to-image generation that uses image representations as pivots, improving prompt translation without needing parallel corpora.

Contribution

PRIP leverages image representations as intermediaries, enabling effective prompt refinement in a zero-shot setting without parallel training data.

Findings

PRIP outperforms various baselines in prompt refinement tasks.

PRIP effectively generalizes to unseen systems in zero-shot scenarios.

Extensive experiments validate the superiority of PRIP over existing methods.

Abstract

For text-to-image generation, automatically refining user-provided natural language prompts into the keyword-enriched prompts favored by systems is essential for the user experience. Such a prompt refinement process is analogous to translating the prompt from "user languages" into "system languages". However, the scarcity of such parallel corpora makes it difficult to train a prompt refinement model. Inspired by zero-shot machine translation techniques, we introduce Prompt Refinement with Image Pivot (PRIP). PRIP innovatively uses the latent representation of a user-preferred image as an intermediary "pivot" between the user and system languages. It decomposes the refinement process into two data-rich tasks: inferring representations of user-preferred images from user languages and subsequently translating image representations into system languages. Thus, it can leverage abundant data for training. Extensive experiments show that PRIP substantially outperforms a wide range of baselines and effectively transfers to unseen systems in a zero-shot manner.