
原文
GPT Image 2 · en
Scene:
Create a square cinematic still for Steam Engine Station Noir. Use production-grade photoreal lighting, disciplined lens behavior, and a composition that feels like a finished key frame from an original film rather than a casual snapshot. A mid-century rail platform is buried in rolling steam and hard-edged station light, with mostly monochrome tonality, wet pavement, iron structure, and one restrained red tail lamp or signal glowing through the vapor. Keep the camera in a disciplined anchor position so the full lighting event and the main spatial structure are immediately legible from edge to edge.
Subject:
Build a coherent original scene whose emotional center comes from atmosphere, light physics, and spatial clarity. Avoid relying on celebrity likeness, franchise iconography, or stylized illustration shortcuts. A lone figure in a trench coat, the locomotive mass, steel beams, and layered steam curtains should build the scene, but the person remains anonymous and secondary to the environment and the light-cutting fog. This frame should lock the canonical arrangement, scale relationship, and emotional tone that the later dependent images inherit.
Important details:
Preserve a believable physical world with strong depth layering, careful color separation, and premium image discipline. Prioritize steam density layering, wet-surface sheen, noir contrast, and restrained color use, so the image feels like a lost still from a serious black-and-white crime film rather than a cosplay set. Use this image to establish the authoritative direction of light, atmospheric density, and structural silhouette for the package.
Use case:
Steam Station series study 1/9.
Constraints:
No readable text, no watermark, no logos, preserve exact 1:1 aspect ratio, no readable station boards, no crowded commuters, no exaggerated detective props, no pristine modern train design, no bright advertising.使い方
- パラメータを含む元のプロンプト全体をコピーします。
- 被写体、文字、構図を自分の要件に置き換えます。
- 対応する外部ツールで使用し、利用条件と結果を確認します。
開始前の確認
- 参考画像の要否を確認してください。互換モデルの表示は、そのモデルでの生成実績を意味しません。
- 名前、数値、画像内の文字を確認します。モデルの版で結果は変わります。





