We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.HC

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Human-Computer Interaction

Title: Best Prompts for Text-to-Image Models and How to Find Them

Abstract: Recent progress in generative models, especially in text-guided diffusion models, has enabled the production of aesthetically-pleasing imagery resembling the works of professional human artists. However, one has to carefully compose the textual description, called the prompt, and augment it with a set of clarifying keywords. Since aesthetics are challenging to evaluate computationally, human feedback is needed to determine the optimal prompt formulation and keyword combination. In this paper, we present a human-in-the-loop approach to learning the most useful combination of prompt keywords using a genetic algorithm. We also show how such an approach can improve the aesthetic appeal of images depicting the same descriptions.
Comments: 12 pages (4 main pages), 4 figures, 4 tables, accepted at NeurIPS 2022 Workshop on Human Evaluation of Generative Models (HEGM)
Subjects: Human-Computer Interaction (cs.HC); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
ACM classes: H.5.2; H.3.3
Cite as: arXiv:2209.11711 [cs.HC]
  (or arXiv:2209.11711v2 [cs.HC] for this version)

Submission history

From: Dmitry Ustalov [view email]
[v1] Fri, 23 Sep 2022 16:39:13 GMT (5143kb,D)
[v2] Fri, 2 Dec 2022 21:44:52 GMT (5143kb,D)

Link back to: arXiv, form interface, contact.