PRISM: Automated Black-box Prompt Engineering for Personalized Text-to-Image Generation

Yutong He1, Alexander Robey1,2, Naoki Murata3, Yiding Jiang1, Joshua Nathaniel Williams1, George J. Pappas2, Hamed Hassani2, Yuki Mitsufuji3,4, Ruslan Salakhutdinov1, J. Zico Kolter1,5,
1Carnegie Mellon University, 2University of Pennsylvania, 3Sony AI, 4Sony Group Corporation, 5Bosch Center for AI
arXiv Paper Code

Given a set of reference images, our method, PRISM, is capable of creating human-interpretable and accurate prompts for the desired concept that are also transferable to both open-sourced and closed-sourced text-to-image models.

Abstract

Prompt engineering is effective for controlling the output of text-to-image (T2I) generative models, but it is also laborious due to the need for manually crafted prompts. This challenge has spurred the development of algorithms for automated prompt generation. However, these methods often struggle with transferability across T2I models, require white-box access to the underlying model, and produce non-intuitive prompts. In this work, we introduce PRISM, an algorithm that automatically identifies human-interpretable and transferable prompts that can effectively generate desired concepts given only black-box access to T2I models. Inspired by large language model (LLM) jailbreaking, PRISM leverages the in-context learning ability of LLMs to iteratively refine the candidate prompts distribution for given reference images. Our experiments demonstrate the versatility and effectiveness of PRISM in generating accurate prompts for objects, styles and images across multiple T2I models, including Stable Diffusion, DALL-E, and Midjourney.

Proposed Method

An illustration of PRISM. The label “System” indicates the system prompts setups for the VLMs.

Demo

Qualitative results for object oriented personalized T2I generation on DreamBooth dataset.

Qualitative results for object oriented personalized T2I generation.

Image inversion results for different methods on different T2I models.

Prompt editing demonstration with Midjourney.

Multi-concept generation demonstration with Midjourney.

BibTeX


        @article{he2024automated,
          title={Automated Black-box Prompt Engineering for Personalized Text-to-Image Generation},
          author={He, Yutong and Robey, Alexander and Murata, Naoki and Jiang, Yiding and Williams, Joshua Nathaniel and Pappas, George J and Hassani, Hamed and Mitsufuji, Yuki and Salakhutdinov, Ruslan and Kolter, J Zico},
          journal={arXiv preprint arXiv:2403.19103},
          year={2024}
        }