画像をyamlでどこまで再現できるか

「画像のサイズやディティールまで含めてyamlプロンプト化してコードボックスに出力して」とうてば再現できます。というSNSを見たので実験。
画像を分析し、そのサイズとディティールを含めたYAMLプロンプトを生成するようだ。

prompt: |
  A portrait photograph of a **young woman** with **long dark hair**, looking directly at the camera with a **serious/neutral expression**.
  She is wearing a **black tank top** or **sleeveless shirt**.
  Her **arms are raised** and her **hands are clasped** or resting on the back of her head, drawing attention to her **shoulders and arms**.
  The lighting is **soft and natural**, possibly **golden hour**, creating a slightly **warm tone** and **subtle highlights** on her skin and hair.
  The background is **softly blurred (bokeh effect)**, consisting of **out-of-focus yellow/green foliage** or a field, suggesting an **outdoor setting**.
  The **focus is sharp** on her face and upper body.
  The image has a **film-like aesthetic** with **fine grain** and **rich, slightly desaturated colors**.
  **Photography style:** 1990s Japanese idol/magazine portrait, natural lighting, high-resolution scan.
aspect_ratio: 3:2
style: cinematic portrait, natural light, soft focus background, film grain
artist: influenced by Kishin Shinoyama, Toru Kitagawa

※artist: influenced by Kishin Shinoyama, Toru Kitagawaは、ChatGPTには却下されました。