Pith. sign in

REVIEW 3 major objections 4 minor 145 references

The paper claims that tracing any image element back to the prompt words that created it — and editing those words rather than the whole prompt — turns trial-and-error AI image refinement into predictable element-level control, with studies

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

GenTune improves AI image refinement by tracing image regions back to prompt labels and allowing element-level, semantic-guided edits.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A competent HCI systems paper with a genuinely useful traceable-prompt interaction loop, but the abstract overstates the stats and the label-mapping component deserves more scrutiny. the 3 major comments →

arxiv 2508.15227 v1 pith:I4PTKFIH submitted 2025-08-21 cs.HC cs.AI

GenTune: Toward Traceable Prompts to Improve Controllability of Image Refinement in Environment Design

Wen-Fan Wang , Ting-Ying Lee , Chien-Ting Lu , Che-Wei Hsu , Nil Ponsa Campany\`a , Yu Chen , Mike Y. Chen , Bing-Yu Chen This is my paper
classification cs.HC cs.AI
keywords generative AIenvironment designtraceable promptsprompt–image mappingimage refinementhuman-centered AIcreativity support toolsemantic-guided refinement
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Environment designers who use generative AI get their initial images from LLM-expanded prompts, but the prompt is opaque: nobody can tell which words produced which element, so refinement becomes trial-and-error, and inpainting repairs one region while breaking the scene's coherence. GenTune claims to fix this with two linked modules — a traceable prompt that maps any selected image region to the label and prompt segment that generated it, and semantic-guided refinement that edits only that label's segment, either regenerating with a controlled seed so the rest of the scene stays stable or writing a context-aware inpainting prompt. In a within-subjects study with 20 designers, GenTune beat a baseline without these modules on prompt-image comprehension, refinement effectiveness, output quality, and satisfaction (p < .01 in each), and a three-day field study at two studios reported faster revisions, better output, and less time explaining changes to clients. If the claim holds, it gives designers a practical middle path between thread-the-needle prompt editing and heavy technical pipelines.

Core claim

GenTune's central claim is that an element-level trace from image back to prompt restores control that current text-to-image workflows lack. The system structures LLM-expanded prompts into six categories (theme, art style, content, lighting, color, shot angle), extracts element labels from the content category, and maps a designer's region selection to the right label by combining SAM segmentation with CLIP semantic similarity. Refinement then operates on that label: a Refinement LLM rewrites the corresponding prompt segment from the user's instruction or reference image and regenerates with a controlled seed so the rest of the scene stays coherent, or an inpainting-prompt LLM produces a sty

What carries the argument

The traceable-prompt pipeline: after a brainstorming LLM expands the user's input into a structured prompt and a label-extraction LLM pulls element labels from the content category, a designer's region selection is segmented by SAM, cropped and alpha-masked, and matched against label texts by CLIP using the template 'The bright part is a segmentation of label'; the winning label reveals its prompt segment. The companion mechanism is semantic-guided refinement: a Refinement LLM edits only the prompt segment bound to that label, then regenerates with the same seed so unchanged content stays visually stable; for local-only edits, an inpainting-prompt LLM writes a context-aware prompt for the in

Load-bearing premise

GenTune's benefit hinges on the label-region mapping being accurate: if the selected region is matched to a wrong, missing, or ambiguous label, the refinement targets the wrong thing, and the paper's own participants report exactly that failure in scenes crowded with similar elements.

What would settle it

Run GenTune's trace stage on scenes with many repeated or overlapping elements (shelves of bottles, rows of similar windows) and compare CLIP's top label against human-annotated region-label pairs; chance-level accuracy on such scenes, or a disappearance of the baseline gap when labels are replaced by human-verified ones, would show the trace premise, rather than the image model, carries the result.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Editing by label changes every element sharing that label in one pass — replacing 'vintage cars' with 'vintage electrical tram' also added overhead wires — collapsing what was a slow multi-region Photoshop task into a single instruction.
  • Seed-controlled prompt regeneration keeps lighting, style, and structure stable where inpainting breaks them, giving designers a refinement option that does not exist in current prompt-and-inpaint practice, and both modes are offered side by side for comparison.
  • The trace-and-revise paradigm carries over to character design, interior design, and game UI, and in principle to video and animation if labels carry attributes like motion and speed.
  • On measured use, GenTune averaged 9.23 minutes and 5.3 iterations per within-subjects task versus 12.50 minutes and 6.8 iterations for the baseline, and the field study reported an estimated 8-hour task finished in 2 hours with three client revision rounds completed in half a day.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The load-bearing assumption is measurable: a benchmark of the SAM-plus-CLIP label mapping against human region-label annotations, weighted toward scenes with repeated or overlapping elements, would locate the failure boundary of the whole approach — the authors' own participants report wrong or ambiguous labels exactly there.
  • Reverse tracing — selecting a label to highlight every image region it governs — is the natural next control; it would have warned the participant whose refinery replaced all 'blackboard'-labeled instances and the one whose shrine refinement also removed the torii gate.
  • As LLM-driven generation becomes more multi-stage (directors, planners, agents), traceability of intermediate outputs becomes a general usability requirement; GenTune is a concrete instance of that principle in images, but the same click-to-intermediate-representation loop could apply to video and code generation.
  • If traceability makes revisions cheap, client-facing iteration volume may rise — the treadmill effect the authors acknowledge — so the efficiency win and the workload shift should be tracked together in longer deployments rather than treating speed as an unqualified gain.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper presents GenTune, a human-centered system for environment designers that makes LLM-expanded text-to-image prompts traceable to image regions and supports semantic-guided refinement (prompt regeneration with a controlled seed or context-aware inpainting). The authors report a formative study with 10 designers, a summative within-subjects experiment and an open-ended task with 20 designers, and a 3-participant field study in two studios. They claim that GenTune significantly improves prompt-image comprehension, refinement effectiveness/quality/efficiency, and overall satisfaction compared to a baseline that lacks the two core modules, with all reported p-values below .01 in the abstract.

Significance. If substantiated, this work addresses a real and timely problem: the opacity of LLM-generated prompts in text-to-image workflows used by environment designers. The contribution is primarily an HCI paradigm—traceable, element-level control—rather than a new generative model. The paper's strengths include a detailed system implementation, a within-subjects comparison against a reasonable baseline that includes conversational editing and basic inpainting, rich qualitative observations, and a field deployment. The self-consistency of the trace labels (labels derive from the prompt that generated the image) is a design feature, not a circularity in the evaluation. However, the evidence base is largely self-reported, the accuracy of the label-region mapping is not quantified despite being load-bearing, and the statistical reporting is internally inconsistent. These issues need to be addressed before the central claims can be accepted as stated.

major comments (3)
  1. [Abstract; §6.1; Fig. 7] The abstract states that GenTune significantly improved prompt-image comprehension, refinement quality and efficiency, and overall satisfaction with 'all p < .01.' In the body, however, §6.1 reports the prompt-image interpretability items (Q1–Q3) only at p < .05, and Fig. 7 marks them with a single asterisk. The blanket 'all p < .01' is therefore contradicted by the presented results. Please correct the abstract and report exact p-values for each item, or qualify the statement appropriately.
  2. [§4.3.1, §6.1, §8.4] The traceable-prompt mechanism is implemented as SAM segmentation followed by CLIP top-5 scoring with a fixed template ('The bright part is a segmentation of label'). The manuscript provides no quantitative accuracy or failure-rate measurement for this mapping. §8.4 concedes that label inaccuracies arise from T2I hallucinations and from an overabundance of similar/ambiguous labels, and §6.1 quotes P11 and P15 reporting wrong or indistinguishable labels in exactly the kind of multi-element scenes environment designers work with. Since RQ1–RQ3 benefits depend on the correctness of the label-region mapping, this is a load-bearing component that is currently unvalidated. Please report mapping accuracy on the study stimuli (e.g., agreement with human-annotated regions) and/or analyze the main outcomes separately for traces that succeeded versus failed. The user study may still support the cla
  3. [§6.2.3 / Table 2; §5.1.3] The abstract claims significantly improved refinement efficiency, but Table 2 reports raw average time and iterations (12.5 vs. 9.23 minutes; 6.8 vs. 5.3 iterations) without any inferential test. The questionnaire items measure perceived efficiency and quality via self-report, and no objective measure of output quality (e.g., independent expert raters blind to condition) is provided. §8.4 acknowledges the reliance on self-report. Please add significance tests for the Table 2 metrics and, ideally, an external evaluation of final image quality, or temper the claims to 'perceived' efficiency and quality. This is important because the 'quality and efficiency' claim is a headline result.
minor comments (4)
  1. [§7] The field study includes only three participants, all of whom were returning collaborators from the earlier studies. This should be described explicitly as an exploratory pilot rather than as evidence of generalizable real-world effectiveness. The manuscript's conclusion ('further demonstrated') is stronger than the design supports.
  2. [§6.1] The phrase '95 and 100% of participants preferred GenTune' is imprecise; please report the actual fractions (e.g., 19/20 and 20/20) for Fig. 8 items Q1–Q3.
  3. [Throughout] Minor typos and formatting issues: 'intial' in §4.3.1, 'instablilty' in §6.3, 'futuraistic' in §7.2, and 'p ¡' should be 'p <' throughout.
  4. [§5.1.3] Multiple Wilcoxon tests are run on the Fig. 7 items without correction or effect sizes. This is acceptable for an exploratory HCI study, but reporting effect sizes and confidence intervals would strengthen the quantitative claims.

Circularity Check

0 steps flagged

No significant circularity; GenTune's claims are grounded in an empirical user study against a baseline, not in a derivation that reduces to its inputs.

full rationale

GenTune is a system paper, not a derivation. The central claims (improved prompt-image comprehension, refinement effectiveness, and satisfaction) come from a within-subjects study comparing GenTune against a baseline that omits the two modules, plus an open-ended task and a field study. The traceable-prompt module extracts labels from the generated structured prompt and uses SAM+CLIP to match selected regions to those labels; this is a design choice about where labels come from, not a fitted parameter later renamed as a prediction. Because the paper does not claim to predict an external ground-truth quantity from first principles, the label-source self-consistency is not circular. The acknowledged label-accuracy limitation in Section 8.4 ('The effectiveness of GenTune's refinement relies heavily on accurate label selection... Label inaccuracies typically arise from hallucinations... and an overabundance of similar or ambiguous labels') is a validity threat to the system's usefulness, but it is an empirical limitation, not a circularity: the user-study outcomes are not logically forced by the label-extraction procedure. The only self-referential element is a citation to the authors' prior AIdeation work [122] as inspiration for the brainstorming module, but that citation is not load-bearing for the paper's main results, which are independently benchmarked against a baseline without the proposed modules. No equation or analysis step reduces to an input by construction, so the appropriate circularity score is 0.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The central user-study claim rests on a set of design decisions and domain assumptions: the label taxonomy, CLIP/SAM accuracy, seed stability, and the validity of self-report. No code or data is released, and none of these assumptions is separately validated with an objective benchmark.

free parameters (3)
  • Outside-mask darkening ratio = 80%
    Chosen in Section 4.3.1 for CLIP input preprocessing; no reported sensitivity analysis.
  • Number of candidate labels returned = 5
    Top-five CLIP similarity labels are returned per region (Section 4.3.1); chosen by hand.
  • Prompt category count = 6
    Theme, art style, content, lighting, color, shot angle; based on formative study but the taxonomy is a design choice.
axioms (5)
  • domain assumption Designers value aesthetic coherence over pixel-level accuracy
    Used to justify seed-based regeneration that may slightly alter non-targeted regions (Sections 3.3.3, 8.1).
  • domain assumption CLIP text-image similarity is a valid proxy for semantic label-region correspondence
    Core to traceable prompt; no validation against ground-truth segmentation labels.
  • domain assumption SAM segmentations accurately capture the user-selected visual element
    Region mask feeds CLIP; inaccurate masks degrade label matching (Section 4.3.1).
  • domain assumption Fixed-seed regeneration with a refined prompt preserves non-targeted elements
    Basis of semantic-guided prompt refinement; paper reports instability counterexamples (P21, Section 6.3).
  • domain assumption Self-reported Likert ratings correspond to actual workflow quality
    All quantitative evidence is self-report; no expert evaluation of output images (Section 8.4).

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of GenTune: Toward Traceable Prompts to Improve Controllability of Image Refinement in Environment Design." pith.science (2026). https://pith.science/paper/I4PTKFIH

@misc{pith2026250815227,
  author       = {Pith},
  title        = {Pith review of: GenTune: Toward Traceable Prompts to Improve Controllability of Image Refinement in Environment Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I4PTKFIH}},
  note         = {Machine review of arXiv:2508.15227}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Environment designers in the entertainment industry create imaginative 2D and 3D scenes for games, films, and television, requiring both fine-grained control of specific details and consistent global coherence. Designers have increasingly integrated generative AI into their workflows, often relying on large language models (LLMs) to expand user prompts for text-to-image generation, then iteratively refining those prompts and applying inpainting. However, our formative study with 10 designers surfaced two key challenges: (1) the lengthy LLM-generated prompts make it difficult to understand and isolate the keywords that must be revised for specific visual elements; and (2) while inpainting supports localized edits, it can struggle with global consistency and correctness. Based on these insights, we present GenTune, an approach that enhances human--AI collaboration by clarifying how AI-generated prompts map to image content. Our GenTune system lets designers select any element in a generated image, trace it back to the corresponding prompt labels, and revise those labels to guide precise yet globally consistent image refinement. In a summative study with 20 designers, GenTune significantly improved prompt--image comprehension, refinement quality, and efficiency, and overall satisfaction (all $p < .01$) compared to current practice. A follow-up field study with two studios further demonstrated its effectiveness in real-world settings.

Figures

Figures reproduced from arXiv: 2508.15227 by Bing-Yu Chen, Che-Wei Hsu, Chien-Ting Lu, Mike Y. Chen, Nil Ponsa Campany\`a, Ting-Ying Lee, Wen-Fan Wang, Yu Chen.

Figure 1
Figure 1. Figure 1: GenTune, a human-centered generative AI system with traceable prompts for controllable [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The main interface of GenTune includes: (A) The main image panel. Users can select [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: GenTune’s refinement workflow from P14 in the open-ended study began with Global [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Prompt-element correspondence pipeline. Label Extraction LLM extracts labels from [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Technical pipeline for semantic-guided prompt refinement and inpainting. If a reference [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Pipeline for three types of suggestion LLM: Global, Label-based, and Expanded sugges [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Responses were collected using a 7-point Likert scale (1 = strongly disagree, 7 = [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Survey results from the within-subject task. Participants rated Prompt-image inter [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Survey results from the open-ended task. User rated their original approach and GenTune [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Workflow procedure and results of using GenTune on real-world projects from two de [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of refinement workflows between the Baseline and GenTune systems for [PITH_FULL_IMAGE:figures/full_fig_p036_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Survey results from the within-subject task. Participants rated the NASA-TLX workload [PITH_FULL_IMAGE:figures/full_fig_p037_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Results from the open-ended task. Each row presents iterative image refinements made [PITH_FULL_IMAGE:figures/full_fig_p037_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Comparison between semantic-guided prompt refinement with controlled seed (middle) [PITH_FULL_IMAGE:figures/full_fig_p038_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison between semantic-guided prompt refinement with controlled seed (middle) [PITH_FULL_IMAGE:figures/full_fig_p038_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Comparison of image refinement results among GenTune, Flux 1.0 (Depth), and [PITH_FULL_IMAGE:figures/full_fig_p039_15.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

145 extracted references · 59 canonical work pages · 8 internal anchors

  1. [1]

    The Ultimate Concept Art Career Guide

    3dtotal Publishing. The Ultimate Concept Art Career Guide . 3dtotal Publishing, 2018

    2018

  2. [2]

    Environment designer

    M3DS Academy. Environment designer. https://www.artstation.com/blogs/ m3dsacademy/zXXz6/environment-designer, 2024

    2024

  3. [3]

    Promptmap: An alternative interaction style for ai- based image generation

    Krzysztof Adamkiewicz, Pawe l Wojciech Wo´ zniak, Julia Dominiak, Andrzej Romanowski, Jakob Karolus, and Stanislav Frolov. Promptmap: An alternative interaction style for ai- based image generation. In Proceedings of the 30th International Conference on Intelligent User Interfaces, IUI ’25, page 1162–1176, New York, NY, USA, 2025. Association for Com- put...

    2025

  4. [4]

    Cognitive processes in iterative design behavior

    Robin S Adams and Cynthia J Atman. Cognitive processes in iterative design behavior. In FIE’99 Frontiers in Education. 29th Annual Frontiers in Education Conference. Designing the Future of Science and Engineering Education. Conference Proceedings (IEEE Cat. No. 99CH37011, volume 1, pages 11A6–13. IEEE, 1999

    1999

  5. [5]

    Crafting effective prompts: A guideline for successful image generation

    Noor Wahyuni Ahmad and Suzana Ruslan. Crafting effective prompts: A guideline for successful image generation. In 2024 14th International Conference on System Engineering and Technology (ICSET), pages 84–89. IEEE, 2024

    2024

  6. [6]

    Prompting for discovery: Flexible sense-making for ai art-making with dreamsheets

    Shm Garanganao Almeda, JD Zamfirescu-Pereira, Kyu Won Kim, Pradeep Mani Rathnam, and Bjoern Hartmann. Prompting for discovery: Flexible sense-making for ai art-making with dreamsheets. In Proceedings of the CHI Conference on Human Factors in Computing Systems, pages 1–17, 2024

    2024

  7. [7]

    The concept of coherence in art

    L Aschenbrenner. The concept of coherence in art. Springer Science & Business Media, 2012

    2012

  8. [8]

    Human-centered ai: The role of human-centered design research in the development of ai

    Jan Auernhammer. Human-centered ai: The role of human-centered design research in the development of ai. 2020

    2020

  9. [9]

    Does the whole exceed its parts? the effect of ai ex- planations on complementary team performance

    Gagan Bansal, Tongshuang Wu, Joyce Zhou, Raymond Fok, Besmira Nushi, Ece Kamar, Marco Tulio Ribeiro, and Daniel Weld. Does the whole exceed its parts? the effect of ai ex- planations on complementary team performance. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems , CHI ’21, New York, NY, USA, 2021. Association for Computi...

    2021

  10. [10]

    A review of ai image generator: influences, challenges, and future prospects for architectural field

    Eleonora Vilgia Putri Beyan, Anastasya Gisela Cinintya Rossy, et al. A review of ai image generator: influences, challenges, and future prospects for architectural field. Journal of Artificial Intelligence in Architecture , 2(1):53–65, 2023

    2023

  11. [11]

    Challenges in generating accurate text in images: A benchmark for text-to-image models on specialized content

    Zenab Bosheah and Vilmos Bilicki. Challenges in generating accurate text in images: A benchmark for text-to-image models on specialized content. Applied Sciences, 15(5):2274, 2025

    2025

  12. [12]

    Generative ai for vision: A comprehensive study of frameworks and applications

    Fouad Bousetouane. Generative ai for vision: A comprehensive study of frameworks and applications. arXiv preprint arXiv:2501.18033 , 2025

    Pith/arXiv arXiv 2025

  13. [13]

    Promp- tify: Text-to-image generation through interactive prompt exploration with large language models

    Stephen Brade, Bryan Wang, Mauricio Sousa, Sageev Oore, and Tovi Grossman. Promp- tify: Text-to-image generation through interactive prompt exploration with large language models. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–14, 2023

    2023

  14. [14]

    The Pocket Mentor for Video Game UX UI (The Pocket Mentors for Games Careers)

    Simon Brewer. The Pocket Mentor for Video Game UX UI (The Pocket Mentors for Games Careers). 1st edition, 2023

    2023

  15. [15]

    Exploring the role of text-to-image ai in concept generation

    Ross Brisco, Laura Hay, and Sam Dhami. Exploring the role of text-to-image ai in concept generation. Proceedings of the Design Society , 3:1835–1844, 2023

    2023

  16. [16]

    Instructpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18392–18402, 2023

    2023

  17. [17]

    Peculiarity of interior design materials for accommodation areas of cruise ships: A state-of-the-art review

    L-S Byun. Peculiarity of interior design materials for accommodation areas of cruise ships: A state-of-the-art review. Ships and Offshore Structures , 1(3):171–183, 2006. 25

    2006

  18. [18]

    Designaid: Using generative ai and semantic di- versity for design inspiration

    Alice Cai, Steven R Rick, Jennifer L Heyman, Yanxia Zhang, Alexandre Filipowicz, Matthew Hong, Matt Klenk, and Thomas Malone. Designaid: Using generative ai and semantic di- versity for design inspiration. In Proceedings of The ACM Collective Intelligence Conference, pages 1–11, 2023

    2023

  19. [19]

    Testing for preference using a sum of wilcoxon signed rank statistics

    Marinela Capanu, Gregory A Jones, and Ronald H Randles. Testing for preference using a sum of wilcoxon signed rank statistics. Computational statistics & data analysis , 51(2):793– 796, 2006

    2006

  20. [20]

    Environment designer

    cgspectrum. Environment designer. https://www.cgspectrum.com/career-pathways/ environment-designer, 2024

    2024

  21. [21]

    Get- ting it right: Improving spatial consistency in text-to-image models

    Agneet Chatterjee, Gabriela Ben Melech Stan, Estelle Aflalo, Sayak Paul, Dhruba Ghosh, Tejas Gokhale, Ludwig Schmidt, Hannaneh Hajishirzi, Vasudev Lal, Chitta Baral, et al. Get- ting it right: Improving spatial consistency in text-to-image models. In European Conference on Computer Vision , pages 204–222. Springer, 2024

    2024

  22. [22]

    Kim, and Nadir Weibel

    Chen Chen, Cuong Nguyen, Thibault Groueix, Vladimir G. Kim, and Nadir Weibel. Memovis: A genai-powered tool for creating companion reference images for 3d design feedback. ACM Trans. Comput.-Hum. Interact., 31(5), November 2024

    2024

  23. [23]

    Autospark: Supporting automobile appearance design ideation with kansei engineering and generative ai

    Liuqing Chen, Qianzhi Jing, Yixin Tsang, Qianyi Wang, Ruocong Liu, Duowei Xia, Yunzhan Zhou, and Lingyun Sun. Autospark: Supporting automobile appearance design ideation with kansei engineering and generative ai. In Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology , pages 1–19, 2024

    2024

  24. [24]

    Towards reasoning era: A survey of long chain-of-thought for reasoning large language models

    Qiguang Chen, Libo Qin, Jinhao Liu, Dengyun Peng, Jiannan Guan, Peng Wang, Mengkang Hu, Yuhang Zhou, Te Gao, and Wangxiang Che. Towards reasoning era: A survey of long chain-of-thought for reasoning large language models. arXiv preprint arXiv:2503.09567, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Next Steps for Human-Centered Generative AI: A Technical Perspective

    Xiang’Anthony’ Chen, Jeff Burke, Ruofei Du, Matthew K Hong, Jennifer Jacobs, Philippe Laban, Dingzeyu Li, Nanyun Peng, Karl DD Willis, Chien-Sheng Wu, et al. Next steps for human-centered generative ai: A technical perspective. arXiv preprint arXiv:2306.15774 , 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  26. [26]

    Elevating and sharpening convergent thinking: The potential of generative ai for creative professionals

    YU-HAN CHIU and Chun-Ching Chen. Elevating and sharpening convergent thinking: The potential of generative ai for creative professionals. Available at SSRN 4911494 , 2024

    2024

  27. [27]

    Cre- ativeconnect: Supporting reference recombination for graphic design ideation with generative ai

    DaEun Choi, Sumin Hong, Jeongeon Park, John Joon Young Chung, and Juho Kim. Cre- ativeconnect: Supporting reference recombination for graphic design ideation with generative ai. In Proceedings of the CHI Conference on Human Factors in Computing Systems , pages 1–25, 2024

    2024

  28. [28]

    Prompting for products: investigating design space exploration strategies for text-to-image generative models

    Leah Chong, I-Ping Lo, Jude Rayan, Steven Dow, Faez Ahmed, and Ioanna Lykourentzou. Prompting for products: investigating design space exploration strategies for text-to-image generative models. Design Science, 11:e2, 2025

    2025

  29. [29]

    Promptpaint: Steering text-to-image generation through paint medium-like interactions

    John Joon Young Chung and Eytan Adar. Promptpaint: Steering text-to-image generation through paint medium-like interactions. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, UIST ’23, New York, NY, USA, 2023. Association for Computing Machinery. 26

    2023

  30. [30]

    The intersection of users, roles, interactions, and technologies in creativity support tools

    John Joon Young Chung, Shiqing He, and Eytan Adar. The intersection of users, roles, interactions, and technologies in creativity support tools. In Proceedings of the 2021 ACM Designing Interactive Systems Conference , DIS ’21, page 1817–1833, New York, NY, USA,

    2021

  31. [31]

    Comfyui: A powerful and modular stable diffusion gui and backend

    ComfyUI Contributors. Comfyui: A powerful and modular stable diffusion gui and backend. https://github.com/comfyanonymous/ComfyUI, 2023

    2023

  32. [32]

    Intellingo: An intelligible translation en- vironment

    Sven Coppers, Jan Van den Bergh, Kris Luyten, Karin Coninx, Iulianna van der Lek-Ciudin, Tom Vanallemeersch, and Vincent Vandeghinste. Intellingo: An intelligible translation en- vironment. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, CHI ’18, page 1–13, New York, NY, USA, 2018. Association for Computing Ma- chinery

    2018

  33. [33]

    Insight-v: Exploring long-chain visual reasoning with multimodal large language models

    Yuhao Dong, Zuyan Liu, Hai-Long Sun, Jingkang Yang, Winston Hu, Yongming Rao, and Ziwei Liu. Insight-v: Exploring long-chain visual reasoning with multimodal large language models. arXiv preprint arXiv:2411.14432 , 2024

    Pith/arXiv arXiv 2024

  34. [34]

    Generative ai enhances individual creativity but reduces the collective diversity of novel content

    Anil R Doshi and Oliver P Hauser. Generative ai enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28):eadn5290, 2024

    2024

  35. [35]

    What is game environment design and how to get started? https://www.domestika.org/en/blog/ 10804-what-is-game-environment-design-and-how-to-get-started , 2022

    Lauren du Plessis. What is game environment design and how to get started? https://www.domestika.org/en/blog/ 10804-what-is-game-environment-design-and-how-to-get-started , 2022

    2022

  36. [36]

    Explainable ai (xai): Core ideas, techniques, and solutions

    Rudresh Dwivedi, Devam Dave, Het Naik, Smiti Singhal, Rana Omer, Pankesh Patel, Bin Qian, Zhenyu Wen, Tejal Shah, Graham Morgan, and Rajiv Ranjan. Explainable ai (xai): Core ideas, techniques, and solutions. ACM Comput. Surv. , 55(9), January 2023

    2023

  37. [37]

    Vera Liao, Elizabeth Anne Watkins, Carina Manger, Hal Daum´ e III, Andreas Riener, and Mark O Riedl

    Upol Ehsan, Philipp Wintersberger, Q. Vera Liao, Elizabeth Anne Watkins, Carina Manger, Hal Daum´ e III, Andreas Riener, and Mark O Riedl. Human-centered explainable ai (hcxai): Beyond opening the black-box of ai. In Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems , CHI EA ’22, New York, NY, USA, 2022. Association for C...

    2022

  38. [38]

    Expertise elevates AI usage: experimental evidence comparing laypeople and professional artists

    Thomas F Eisenmann, Andres Karjus, Mar Canet Sola, Levin Brinkmann, Braman- tyo Ibrahim Supriyatno, and Iyad Rahwan. Expertise elevates ai usage: experimental evidence comparing laypeople and professional artists. arXiv preprint arXiv:2501.12374 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    Ganzilla: User-driven direction discovery in gen- erative adversarial networks

    Noyan Evirgen and Xiang’Anthony’ Chen. Ganzilla: User-driven direction discovery in gen- erative adversarial networks. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology, pages 1–10, 2022

    2022

  40. [40]

    Ganravel: User-driven direction disentanglement in generative adversarial networks

    Noyan Evirgen and Xiang’Anthony Chen. Ganravel: User-driven direction disentanglement in generative adversarial networks. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems , pages 1–15, 2023

    2023

  41. [41]

    From text to pixels: Enhancing user understanding through text-to-image model explanations

    Noyan Evirgen, Ruolin Wang, and Xiang ’Anthony Chen. From text to pixels: Enhancing user understanding through text-to-image model explanations. In Proceedings of the 29th International Conference on Intelligent User Interfaces , IUI ’24, page 74–87, New York, NY, USA, 2024. Association for Computing Machinery. 27

    2024

  42. [42]

    Promptmagician: Interactive prompt engineering for text-to- image creation

    Yingchaojie Feng, Xingbo Wang, Kam Kwai Wong, Sijia Wang, Yuhong Lu, Minfeng Zhu, Baicheng Wang, and Wei Chen. Promptmagician: Interactive prompt engineering for text-to- image creation. IEEE Transactions on Visualization and Computer Graphics , 30(1):295–305, 2023

    2023

  43. [43]

    How digital tools im- pact convergent and divergent thinking in design ideation

    Jonas Frich, Midas Nouwens, Kim Halskov, and Peter Dalsgaard. How digital tools im- pact convergent and divergent thinking in design ideation. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems , CHI ’21, New York, NY, USA, 2021. Association for Computing Machinery

    2021

  44. [44]

    "I Like Your Story!": A Co-Creative Story-Crafting Game with a Persona-Driven Character Based on Generative AI

    Jiaying Fu, Xiruo Wang, Zhouyi Li, Kate Vi, Chuyan Xu, and Yuqian Sun. ” i like your story!”: A co-creative story-crafting game with a persona-driven character based on generative ai. arXiv preprint arXiv:2503.09102 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  45. [45]

    The paradox of artificial creativity: Challenges and opportunities of gen- erative ai artistry

    Manuel B Garcia. The paradox of artificial creativity: Challenges and opportunities of gen- erative ai artistry. Creativity Research Journal, pages 1–14, 2024

    2024

  46. [46]

    Seed-data-edit technical report: A hybrid dataset for instructional image editing

    Yuying Ge, Sijie Zhao, Chen Li, Yixiao Ge, and Ying Shan. Seed-data-edit technical report: A hybrid dataset for instructional image editing. arXiv preprint arXiv:2405.04007 , 2024

    Pith/arXiv arXiv 2024

  47. [47]

    Prompthis: Visualizing the process and influence of prompt editing during text-to-image creation

    Yuhan Guo, Hanning Shao, Can Liu, Kai Xu, and Xiaoru Yuan. Prompthis: Visualizing the process and influence of prompt editing during text-to-image creation. IEEE Transactions on Visualization and Computer Graphics , 2024

    2024

  48. [48]

    Archcode: In- corporating software requirements in code generation with large language models

    Hojae Han, Jaejin Kim, Jaeseok Yoo, Youngwon Lee, and Seung-won Hwang. Archcode: In- corporating software requirements in code generation with large language models. In Proceed- ings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages 13520–13552, 2024

    2024

  49. [49]

    Multi-stage guided code generation for large language models

    Yewei Han and Chen Lyu. Multi-stage guided code generation for large language models. Engineering Applications of Artificial Intelligence , 139:109491, 2025

    2025

  50. [50]

    Viewdiff: 3d-consistent image gener- ation with text-to-image models

    Lukas H¨ ollein, Aljaˇ z Boˇ ziˇ c, Norman M¨ uller, David Novotny, Hung-Yu Tseng, Christian Richardt, Michael Zollh¨ ofer, and Matthias Nießner. Viewdiff: 3d-consistent image gener- ation with text-to-image models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5043–5052, 2024

    2024

  51. [51]

    Di- rect2v: Large language models are frame-level directors for zero-shot text-to-video generation

    Susung Hong, Junyoung Seo, Heeseong Shin, Sunghwan Hong, and Seungryong Kim. Di- rect2v: Large language models are frame-level directors for zero-shot text-to-video generation. arXiv preprint arXiv:2305.14330 , 2023

    Pith/arXiv arXiv 2023

  52. [52]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2021

    2021

  53. [53]

    Dialoggen: Multi-modal interactive dialogue system for multi-turn text-to-image generation

    Minbin Huang, Yanxin Long, Xinchi Deng, Ruihang Chu, Jiangfeng Xiong, Xiaodan Liang, Hong Cheng, Qinglin Lu, and Wei Liu. Dialoggen: Multi-modal interactive dialogue system for multi-turn text-to-image generation. arXiv preprint arXiv:2403.08857 , 2024

    Pith/arXiv arXiv 2024

  54. [54]

    Plantography: Incorporating iterative design process into generative artificial intelligence for landscape ren- dering

    Rong Huang, Haichuan Lin, Chuanzhang Chen, Kang Zhang, and Wei Zeng. Plantography: Incorporating iterative design process into generative artificial intelligence for landscape ren- dering. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, New York, NY, USA, 2024. Association for Computing Machinery. 28

    2024

  55. [55]

    Videodiff: Human-ai video co-creation with alternatives

    Mina Huh, Dingzeyu Li, Kim Pimmel, Hijung Valentina Shin, Amy Pavel, and Mira Dontcheva. Videodiff: Human-ai video co-creation with alternatives. arXiv preprint arXiv:2502.10190, 2025

    Pith/arXiv arXiv 2025

  56. [56]

    The association of creativity with diver- gent and convergent thinking

    Syed Fahad Javaid and James Paul Pandarakalam. The association of creativity with diver- gent and convergent thinking. Psychiatria danubina , 33(2):133–139, 2021

    2021

  57. [57]

    Shih, and Kyungsik Han

    Youngseung Jeon, Seungwan Jin, Patrick C. Shih, and Kyungsik Han. Fashionq: An ai-driven creativity support tool for facilitating ideation in fashion design. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems , CHI ’21, New York, NY, USA,

    2021

  58. [58]

    Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Work- man, Alex Hanna, Johnathan Flowers, and Timnit Gebru

    Harry H. Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Work- man, Alex Hanna, Johnathan Flowers, and Timnit Gebru. Ai art and its impact on artists. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society , AIES ’23, page 363–374, New York, NY, USA, 2023. Association for Computing Machinery

    2023

  59. [60]

    Analysis and refinement of iterative design processes

    Eric William Johnson. Analysis and refinement of iterative design processes . University of Notre Dame, 1997

    1997

  60. [61]

    Self-planning code generation with large language models

    Xue Jiang, Yihong Dong, Lecheng Wang, Zheng Fang, Qiwei Shang, Ge Li, Zhi Jin, and Wenpin Jiao. Self-planning code generation with large language models. ACM Transactions on Software Engineering and Methodology , 33(7):1–30, 2024

    2024

  61. [62]

    Sunnie S. Y. Kim, Elizabeth Anne Watkins, Olga Russakovsky, Ruth Fong, and Andr´ es Monroy-Hern´ andez. ”help me help the ai”: Understanding how explainability can support human-ai interaction. In Proceedings of the 2023 CHI Conference on Human Factors in Com- puting Systems , CHI ’23, New York, NY, USA, 2023. Association for Computing Machinery

    2023

  62. [63]

    Convergent versus divergent thinking

    Kyung Hee Kim and Robert A Pierce. Convergent versus divergent thinking. Encyclopedia of creativity, invention, innovation and entrepreneurship , pages 245–250, 2013

    2013

  63. [64]

    Large- scale text-to-image generation models for visual artists’ creative works

    Hyung-Kwon Ko, Gwanmo Park, Hyeon Jeon, Jaemin Jo, Juho Kim, and Jinwook Seo. Large- scale text-to-image generation models for visual artists’ creative works. In Proceedings of the 28th international conference on intelligent user interfaces , pages 919–933, 2023

    2023

  64. [65]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4015–4026, 2023

    2023

  65. [66]

    Human creativ- ity in the age of llms: Randomized experiments on divergent and convergent thinking

    Harsh Kumar, Jonathan Vincentius, Ewan Jordan, and Ashton Anderson. Human creativ- ity in the age of llms: Randomized experiments on divergent and convergent thinking. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , CHI ’25, New York, NY, USA, 2025. Association for Computing Machinery

    2025

  66. [67]

    Un- derstanding visual artists’ adoption of large-scale text-to-image generation models for creative works

    Hyung-Kwon Ko, Gwanmo Park, Hyeon Jeon, Jaemin Jo, Juho Kim, and Jinwook Seo. Un- derstanding visual artists’ adoption of large-scale text-to-image generation models for creative works. In Proceedings of the 2023 ACM SIGCHI Conference on Human Factors in Computing Systems, pages 1–13, 2023

    2023

  67. [68]

    Flowgpt: Exploring domains, output modalities, and goals of community-generated ai chatbots

    Xian Li, Yuanning Han, Di Liu, Pengcheng An, and Shuo Niu. Flowgpt: Exploring domains, output modalities, and goals of community-generated ai chatbots. In Companion Publication of the 2024 Conference on Computer-Supported Cooperative Work and Social Computing , pages 355–361, 2024

    2024

  68. [69]

    Enhancing compositional text-to- image generation with reliable random seeds

    Shuangqi Li, Hieu Le, Jingyi Xu, and Mathieu Salzmann. Enhancing compositional text-to- image generation with reliable random seeds. arXiv preprint arXiv:2411.18810 , 2024. 29

    Pith/arXiv arXiv 2024

  69. [70]

    Eliott J. Lilly. Big Bad World of Concept Art for Video Games: An Insider’s Guide for Students. Design Studio Press, 2015

    2015

  70. [71]

    Human-centered explainable ai (xai): From algorithms to user experiences

    Q Vera Liao and Kush R Varshney. Human-centered explainable ai (xai): From algorithms to user experiences. arXiv preprint arXiv:2110.10790 , 2021

    Pith/arXiv arXiv 2021

  71. [72]

    SketchFlex: Facilitating Spatial-Semantic Coherence in Text-to-Image Generation with Region-Based Sketches

    Haichuan Lin, Yilin Ye, Jiazhi Xia, and Wei Zeng. Sketchflex: Facilitating spatial- semantic coherence in text-to-image generation with region-based sketches. arXiv preprint arXiv:2502.07556, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  72. [73]

    Inkspire: Supporting Design Exploration with Generative AI through Analogical Sketching

    David Chuan-En Lin, Hyeonsu B Kang, Nikolas Martelaro, Aniket Kittur, Yan-Ying Chen, and Matthew K Hong. Inkspire: Supporting design exploration with generative ai through analogical sketching. arXiv preprint arXiv:2501.18588 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  73. [74]

    Design guidelines for prompt engineering text-to-image gen- erative models

    Vivian Liu and Lydia B Chilton. Design guidelines for prompt engineering text-to-image gen- erative models. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems, CHI ’22, New York, NY, USA, 2022. Association for Computing Machinery

    2022

  74. [75]

    From prompt engineering to prompt craft

    Joseph Lindley and Roger Whitham. From prompt engineering to prompt craft. In Pro- ceedings of the Nineteenth International Conference on Tangible, Embedded, and Embodied Interaction, pages 1–12, 2025

    2025

  75. [76]

    3dall-e: Integrating text- to-image ai in 3d design workflows

    Vivian Liu, Jo Vermeulen, George Fitzmaurice, and Justin Matejka. 3dall-e: Integrating text- to-image ai in 3d design workflows. In Proceedings of the 2023 ACM Designing Interactive Systems Conference, DIS ’23, page 1955–1977, New York, NY, USA, 2023. Association for Computing Machinery

    2023

  76. [77]

    Opal: Multimodal image generation for news illustration

    Vivian Liu, Han Qiao, and Lydia Chilton. Opal: Multimodal image generation for news illustration. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology, UIST ’22, New York, NY, USA, 2022. Association for Computing Machinery

    2022

  77. [78]

    Elaborating a framework that is able to structure and evaluate design workflow and composition of generative ai visualiza- tions

    Gianmarco Longo, Deborah Middleton, and Silvia Albano. Elaborating a framework that is able to structure and evaluate design workflow and composition of generative ai visualiza- tions. In IHIET-AI 2024: 11th International Conference on Human Interaction & Emerging Technologies: Artificial Intelligence & Future Applications. AHFE International Open Access, 2024

    2024

  78. [79]

    Sketchar: Supporting character design and illustration prototyping using generative ai

    LING Long, CHEN Xinyi, WEN Ruoyu, LI Toby Jia-Jun, and LC Ray. Sketchar: Supporting character design and illustration prototyping using generative ai. Proceedings of the ACM on Human-Computer Interaction, 8(CHI PLAY):337, 2024

    2024

  79. [80]

    Llm-generated explanations for recommender systems

    Sebastian Lubos, Thi Ngoc Trang Tran, Alexander Felfernig, Seda Polat Erdeniz, and Viet- Man Le. Llm-generated explanations for recommender systems. In Adjunct Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization , pages 276–285, 2024

    work page 2024

  80. [81]

    Ryan Louie, Andy Coenen, Cheng Zhi Huang, Michael Terry, and Carrie J. Cai. Novice-ai music co-creation via ai-steering tools for deep generative models. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems , CHI ’20, page 1–13, New York, NY, USA, 2020. Association for Computing Machinery. 30

    work page 2020

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.