Prompt Recovery for Image Generation Models: A Comparative Study of Discrete Optimizers
Keywords: Discrete Optimization, Prompt Inversion, Benchmarking
TL;DR: We benchmark several approaches to prompt inversion for image generation models with a particular focus on methods of discrete optimization of embeddings
Abstract: Recovering natural language prompts for image generation models, solely based on the generated images is a difficult discrete optimization problem. In this work, we present the first head-to-head comparison of recent discrete optimization techniques for the problem of prompt inversion. Following prior work on prompt inversion, we use CLIP's (Radford et al., 2021) text-image alignment as an inexpensive proxy for the distribution of prompt-image pairs, and compare several discrete optimizers against BLIP2's image captioner (Li et al., 2024) and PRISM (He et al., 2024) in order to evaluate the quality of discretely optimized prompts across various metrics related to the quality of inverted prompts and the images that they generate. We find that while the discrete optimizers effectively minimize their objectives, CLIP similarity between the inverted prompts and the ground truth image acts as a poor proxy for the distribution of prompt-image pairs -- responses from well-trained captioners often lead to generated images that more closely resemble those produced by the original prompts. This finding highlights the need for further investigation into inexpensive methods of modeling the relationship between the prompts for generative models and their output space.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7782
Loading