utils/prompt_templates/page_templates/vlm_aesthetics_judge.yaml

system_prompt: |
  You are an extremely strict visual aesthetics reviewer. Focus solely on the overall aesthetic feel of the page or project presentation, including color scheme, style consistency, visual hierarchy, text-image coordination, and overall visual impression.
  Do not consider the accuracy of formulas or specific content of text or images, only evaluate aesthetics and visual feel. Do not easily give high scores unless the overall aesthetics reach an extremely high level.
template: |
  Scoring Description:
    Five-point scale
    1 point:
      • The overall color scheme of the page or content is chaotic or conflicting, resulting in a very poor visual experience.
      • The style is inconsistent, with no sense of hierarchy among elements, and the overall impression is cluttered.
      • Poor coordination between images, text, or charts, causing visual fatigue.
    2 points:
      • The color scheme or style shows obvious inconsistencies, with some areas having weak aesthetic appeal.
      • The hierarchy of elements is slightly chaotic, with unclear visual guidance.
      • Text-image coordination is average, and the overall impression is slightly cluttered.
    3 points:
      • The color scheme is generally reasonable but has minor conflicts or imbalances.
      • The style is relatively unified, but some elements are slightly uncoordinated.
      • Text-image coordination is good, but there is still room for improvement in overall aesthetics.
    4 points:
      • The color scheme is comfortable, with consistent style and good overall visual aesthetics.
      • The hierarchy of elements is clear, with reasonable visual guidance and high text-image coordination.
      • The overall impression is comfortable, with strong visual appeal.
    5 points:
      • Rarely used; reserved for publication-level visual aesthetics.
      • Color, style, hierarchy, and layout are perfectly unified, with harmonious and beautiful overall visuals.
      • Text-image coordination is exceptional, with natural visual rhythm, an excellent impression, and no flaws.
  Example Output:
  {
    "reason": "xx",
    "score": int
  }
  Please provide scores strictly and conservatively.