CoChat
Sign Up
Cited by 124

The DOI matches its Crossref record; 23 of the 23 references it cites were found in a registry or an index.

PromptMagician: Interactive Prompt Engineering for Text-to-Image Creation

Feng, Yingchaojie, Wang, Xingbo, Wong, Kam Kwai, Wang, Sijia, Lu, Yuhong, Zhu, Minfeng, Wang, Baicheng, Chen, Wei·2023·IEEE Transactions on Visualization and Computer Graphics
DOI 10.1109/tvcg.2023.3327168
Open this paper in CoChat — free account

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1109/tvcg.2023.3327168

Cites

23 of 23 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 The user interface of PromptMagician consists of four views. The Model Input View (A) configures the prompts and hyperparameters for image creation. The Image Browser View (B) visualizes the generated and retrieved images and the recommended prompt keywords. The Image Evaluation View (C) helps evaluate and filter images based on multiple criteria. The Local Exploration View (D) helps users explore and validate the prompt keywords and guidance scales for images of interest.
Fig. 1 The user interface of PromptMagician consists of four views. The Model Input View (A) configures the prompts and hyperparameters for image creation. The Image Browser View (B) visualizes the generated and retrieved images and the recommended prompt keywords. The Image Evaluation View (C) helps evaluate and filter images based on multiple criteria. The Local Exploration View (D) helps users explore and validate the prompt keywords and guidance scales for images of interest.CC-BY · From the PDF
Fig. 2 The PromptMagician framework consists of four major components. It enables users to (A) specify model input for text-to-image creation. PromptMagician (B) generates a set of images using Stable Diffusion and (C) identifies related prompt keywords for recommendations. (D) Both the image results and prompt keywords are visualized in the user interface to support interactive exploration for prompt engineering.
Fig. 2 The PromptMagician framework consists of four major components. It enables users to (A) specify model input for text-to-image creation. PromptMagician (B) generates a set of images using Stable Diffusion and (C) identifies related prompt keywords for recommendations. (D) Both the image results and prompt keywords are visualized in the user interface to support interactive exploration for prompt engineering.CC-BY · From the PDF
Fig. 3 The pipeline of the prompt recommendation model involves five steps: (A) retrieving similar images from the DiffusionDB dataset; (B) embedding them according to their semantics; (C) arranging them into hierarchical clusters; (D) mining special and related keywords from the prompts in the clusters; and (E) matching each keyword with its most related cluster.
Fig. 3 The pipeline of the prompt recommendation model involves five steps: (A) retrieving similar images from the DiffusionDB dataset; (B) embedding them according to their semantics; (C) arranging them into hierarchical clusters; (D) mining special and related keywords from the prompts in the clusters; and (E) matching each keyword with its most related cluster.CC-BY · From the PDF
Fig. 4 The first usage scenario showcases how our system supports the user in refining the prompt for the desired image style. (A) The user tries the first prompt but is not satisfied with the model-generated images. (B) She then explores similar images and receives recommended prompt keywords that clarify the desired image style. (C) Finally, she successfully obtains the desired results on the second attempt.
Fig. 4 The first usage scenario showcases how our system supports the user in refining the prompt for the desired image style. (A) The user tries the first prompt but is not satisfied with the model-generated images. (B) She then explores similar images and receives recommended prompt keywords that clarify the desired image style. (C) Finally, she successfully obtains the desired results on the second attempt.CC-BY · From the PDF
Fig. 5 The second usage scenario presents how our system facilitates open-ended creation. (A) The user inputs the first prompt and gets unexpected image results. (B) So he overviews the image collection and explores the prompt keywords for city and robot images separately. (C) Then he revises the original prompt and gets a new collection of images. (D) To improve the details of the images, the user specifies criteria to select highly detailed images. (E) With the recommended keywords, the user finally gets a satisfied image result.
Fig. 5 The second usage scenario presents how our system facilitates open-ended creation. (A) The user inputs the first prompt and gets unexpected image results. (B) So he overviews the image collection and explores the prompt keywords for city and robot images separately. (C) Then he revises the original prompt and gets a new collection of images. (D) To improve the details of the images, the user specifies criteria to select highly detailed images. (E) With the recommended keywords, the user finally gets a satisfied image result.CC-BY · From the PDF
Fig. 6 The results of the questionnaire regarding the effectiveness and usability of the visual system and prompt recommendation model.
Fig. 6 The results of the questionnaire regarding the effectiveness and usability of the visual system and prompt recommendation model.CC-BY · From the PDF
Fig. 7 The results of the questionnaire regarding the creativity support of our system and two baseline systems.
Fig. 7 The results of the questionnaire regarding the creativity support of our system and two baseline systems.CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 23 references

Research works printed but not linked: 60

  1. 10.1109/tvcg.2022.3209425 · DOI
  2. 10.1109/tvcg.2006.187 · DOI
  3. Chambon, 2022, Adapting pretrained vision-language foundational models to medical imaging domains, arXiv
  4. 10.1109/tvcg.2021.3138933 · DOI
  5. 10.1145/3491102.3517485 · DOI
  6. 10.1109/cvprw53098.2021.00054 · DOI
  7. 10.1007/s12650-021-00780-0 · DOI
  8. 10.1109/tvcg.2023.3240003 · DOI
  9. 10.1145/2807442.2807478 · DOI
  10. Hao, 2022, Optimizing prompts for text-to-image generation, arXiv
  11. 10.1109/TVCG.2023.3326586 · DOI
  12. Ho, 2022, Imagen video: High definition video generation with diffusion models, arXiv
  13. Liang, 2022, Multiviz: Towards visualizing and understanding multimodal models
  14. 10.1109/tvcg.2020.3030370 · DOI
  15. 10.1145/3526113.3545621 · DOI · Paper page
  16. 10.1186/1471-2105-13-275 · DOI
  17. 10.3115/1118108.1118117 · DOI
  18. Luca, 2019, Multi-level tree based approach for interactive graph visualization with semantic zoom, arXiv
  19. 10.1109/tvcg.2021.3114876 · DOI
  20. 10.1016/j.cviu.2016.12.009 · DOI
  21. MacQueen, Classification and analysis of multivariate observations, 281
  22. 10.1109/mra.2012.2192811 · DOI · Paper page
  23. Nichol, Glide: Towards photorealistic image generation and editing with text-guided diffusion models, 162, 16784
  24. 2023, GPT-4 technical report, arXiv, 2303
  25. Oppenlaender, 2022, Prompt engineering for text-based generative art, arXiv
  26. 10.1080/0144929x.2023.2286532 · DOI
  27. 10.1145/3588432.3591500 · DOI
  28. 10.1007/978-3-030-01216-8_36 · DOI
  29. 10.1145/166117.166125 · DOI
  30. 10.18653/v1/2021.naacl-main.208 · DOI · Paper page
  31. 10.18653/v1/2021.naacl-main.185 · DOI · Paper page
  32. 10.18653/v1/2020.emnlp-main.346 · DOI · Paper page
  33. Song, 2021, Denoising diffusion implicit models
  34. 10.1108/00220410410560573 · DOI
  35. 10.1109/mcg.2020.2986902 · DOI
  36. Van der Maaten, 2008, Visualizing data using t-sne, Journal of machine learning research, 9
  37. 10.1609/aaai.v37i2.25353 · DOI
  38. Wang, 2022, Interactive data analysis with next-step natural language query recommendation, arXiv
  39. 10.1109/tvcg.2021.3114794 · DOI
  40. 10.1007/s11704-023-2691-y · DOI
  41. 10.18653/v1/2023.acl-long.51 · DOI
  42. Weng, 2023, Towards an understanding and explanation for mixed-initiative artificial scientific text detection, arXiv
  43. 10.1109/tvcg.2023.3245609 · DOI
  44. 10.1145/3491101.3519729 · DOI · Paper page
  45. 10.1145/3491102.3517582 · DOI
  46. 10.1109/tvcg.2022.3209423 · DOI
  47. 10.1109/TVCG.2018.2835485 · DOI
  48. 10.1109/VAST.2006.261425 · DOI
  49. 10.1109/tvcg.2021.3128157 · DOI
  50. 10.1109/cvpr.2012.6247789 · DOI
  51. 10.1109/tvcg.2022.3169175 · DOI
  52. 10.1109/tvcg.2019.2934656 · DOI
  53. 10.1109/iccv.2017.629 · DOI · Paper page
  54. 10.1109/tip.2011.2109730 · DOI
  55. Zhang, 2023, Scrolltimes: Tracing the provenance of paintings as a window into history, arXiv
  56. 10.1109/tvcg.2022.3209483 · DOI
  57. Zhang, 2023, Computational approaches for traditional chinese painting: From the “six principles of painting” perspective, arXiv
  58. 10.1109/tvcg.2022.3209391 · DOI
  59. 10.1016/j.visinf.2021.06.002 · DOI
  60. 10.1109/cvpr.2019.00595 · DOI

Cited by

Sorted by citations

Metadata

Authors
Feng, Y., Wang, X., Wong, K. K., Wang, S., Lu, Y., Zhu, M., Wang, B., & Chen, W.
Affiliations
  • State Key Lab of CAD&CG, Zhejiang University, China
  • Hong Kong University of Science and Technology, Hong Kong
  • Zhejiang University, China
ORCID
Venue
IEEE Transactions on Visualization and Computer Graphics
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2023
Pages
1-11
ISSN
  • 1077-2626
  • 1941-0506
  • 2160-9306
DOI
10.1109/tvcg.2023.3327168
Cited by
124
References
23
Licence

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review