The article presents a new framework called Knowledge Pursuit Prompting (KPP) designed to improve the quality of text-driven generative models. The proposed framework iteratively incorporates external knowledge to help these models produce reliable visual content. This is achieved through a recursive process that gathers informative facts from a knowledge base, instructs a language model to compress the acquired knowledge for prompt refinement, and then uses the refined prompts for visual synthesis. The authors evaluate the KPP framework across multiple generative tasks and demonstrate that it is capable of generating faithful and semantically rich content across diverse visual domains.
Publication date: 30 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.17898
