Visual Discomfort in the Built Environment: Leveraging Generative AI and Computational Analysis to Evaluate Predicted Visual Stress in Architectural Façades

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study describes an integrated analytical methodology that merges generative artificial intelligence to measure the complexity of architectural facades that may contribute to visual disturbance. After the first page, the use of sources in the article is extremely inadequate and it has been determined that one of the authors' works is cited. Insufficient references have been given for many findings. In particular, the explanations in the title Current Methodological Approaches to Measuring Visual Stress in Architectural Environments are not supported by the literature. Table 6 and Figure 1 are not readable. It should be explained what type of data the artificial intelligence was trained with and what kind of algorithm was used. In addition, information should be provided about the data used to train the model and the dataset analyzed. A graphical narrative is needed to understand the model. The results should be explained in relation to the visuals used.
Its impact on architecture should be discussed.

Author Response

Please see response attached. 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Author,

Thank you for good paper. I have listed the items below as important issues.

Scientific Evaluation

Strengths

- The integration of generative AI (Midjourney) with Fourier-based computational analysis to assess architectural façades is innovative and highly relevant.

- The paper addresses a critical gap in neuroarchitectural research by proposing an interdisciplinary framework combining architecture, visual neuroscience, and artificial intelligence.

Areas for Improvement

a. Lack of Human-Subject Validation:

The study relies entirely on computational tools (ViStA), without empirical validation through physiological or psychological testing (e.g., fNIRS, EEG, subjective discomfort ratings). Incorporating human data would significantly strengthen the conclusions.

b. Limited Sample Diversity:

Only 9 façade designs are analyzed. A broader set of variations would improve the robustness and generalizability of the findings.

c. Insufficient Control Over Environmental Factors:

While lighting and contrast are mentioned, real-world conditions (e.g., time of day, ambient lighting, reflectivity) are not systematically controlled or discussed.

d. Population-Specific Sensitivity Not Considered:

The paper emphasizes neurodiverse groups’ susceptibility but does not empirically evaluate this subgroup. Including participant testing or referencing studies with these demographics would increase depth.

Language and Structure Evaluation

Strengths

- The manuscript demonstrates a high level of academic language and familiarity with interdisciplinary terminology.

- The writing is clear, professional, and rich in technical vocabulary, appropriate for a scholarly audience.

Issues and Recommendations

a. Long and Complex Sentences:

Many sentences are overly long, containing multiple clauses that reduce readability. Breaking them into two or more sentences would enhance clarity.

b. Repetition of Technical Terms:

Terms such as “visual stress,” “ecological validity,” and “spatial frequency” are repeated excessively. Varying expressions or using referential terms can improve fluency.

c. Paragraph Density:

Several paragraphs exceed 20 lines without breaks, making it hard to digest. Consider breaking them up to guide the reader through complex ideas.

d. Vague Referential Phrases:

The use of words like “this” or “these” without a clear antecedent can be confusing in academic texts. Ensure all references are explicitly tied to the preceding idea.

Recommendations

- Consider a professional English language editing pass to improve clarity and reduce redundancy.

- Add experimental data or survey-based validation to support computational predictions.

• Expand the dataset of façade samples to encompass more stylistic and spatial variability.
•  

 

 

 

 

 

 

Author Response

Please see response attached. 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Check the attached file.

Comments for author File: Comments.pdf

Author Response

Please see response attached. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The article can be published in this form.

Author Response

Thank you!

Reviewer 3 Report

Comments and Suggestions for Authors

Visual Discomfort in the Built Environment: Leveraging Generative AI and Computational Analysis to Evaluate Predicted Visual Stress in Architectural Façades

 

My comments and suggestions are the following:

 

1. Your computational ViStA tool assumes that deviations from natural image statistics directly translate into cortical hyperexcitability and visual stress. Provide rigorous neurophysiological validation using at least three independent modalities (e.g., fNIRS, EEG, HRV) in a controlled human-subject study that empirically correlates ViStA metrics with measurable cortical and autonomic responses across neurotypical and neurodivergent populations.

 

2. Your methodology is based on static 2D façade images. However, architectural experience is inherently dynamic and multisensory. Design and execute an immersive, time-resolved experimental paradigm (e.g., using VR or AR environments) to assess how movement, peripheral vision, and temporal exposure modulate visual stress metrics over time. Recalibrate your ViStA tool to accommodate dynamic stimuli.

 

3. Midjourney is a non-transparent, proprietary black-box system. You claim controlled generative design, but cannot verify latent variables. Rebuild your façade generation pipeline using an open-source diffusion-based architecture (e.g., Stable Diffusion or ControlNet) with full prompt control, code transparency, and pixel-space constraint logic. Retrain the model with architectural datasets to enable reproducibility and scientific scrutiny.

 

4. All façade variants are derived from a Western architectural typology. Conduct a cross-cultural generalizability study by adapting your experimental framework to diverse urban contexts (e.g., East Asian, African, Middle Eastern architectural paradigms). Determine whether visual stress metrics and critical spatial frequencies hold across climatically and culturally divergent design principles.

 

5. Your generative prompts manipulate multiple façade parameters simultaneously (contrast, spatial frequency, repetition). Develop a factorial design experiment where each visual attribute is isolated as an independent variable and tested across all possible permutations. Quantify and rank each factor’s marginal contribution to ViStA metrics using interpretable machine learning (e.g., SHAP values, hierarchical Bayesian modeling).

 

6. You link visual stress to allostatic load without longitudinal physiological evidence. Design a month/year-long cohort study where participants are exposed daily to varied architectural visual stimuli (real or VR), and measure cumulative stress markers (e.g., cortisol levels, inflammatory cytokines, HRV entropy). Show how your tool’s outputs correlate with progressive physiological dysregulation.

 

7. Visual stress susceptibility varies significantly across individuals. Extend your framework with a neurodiversity-aware module trained on individual fMRI/EEG responses. Implement adaptive visual stress prediction models (e.g., personalized ViStA tuning) that account for conditions like autism, PTSD, or sensory processing disorder. Validate this with stratified neurocognitive testing.

 

8. Your study isolates visual factors, but real environments involve multimodal inputs (acoustics, temperature, crowd density). Integrate multimodal environmental stressors into your analytical pipeline and determine interaction effects on visual discomfort. Use multisensory simulation chambers or mixed-reality setups to generate interaction-aware visual stress maps.

 

9. Your visual stress model relies entirely on Fourier transforms and 1/f deviation. Explore alternative image representations grounded in biologically plausible vision systems (e.g., wavelet decomposition, Gabor filters, retinal ganglion cell modeling). Compare their predictive power against ViStA using empirical discomfort ratings and determine which best maps to cortical excitation patterns.

 

10. Assuming your tool is integrated into large-scale urban planning workflows, propose and validate an ethical framework and decision-support system that can classify city-wide visual stress zones. Use satellite or drone-based façade imagery from multiple global cities. Develop a prototype GIS-based interface that highlights visual stress clusters and simulates intervention outcomes.

 

Author Response

Please see attached document

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

Comments and Suggestions for Authors

Accept