
原始文档
GPT Image 2 · en
Scene:
Create a square bust-focused crop designed like a premium character-select portrait. The camera should frame from upper chest to just above the head, with slight three-quarter rotation, controlled studio lighting, and a background made of layered color blocks, subtle faction geometry, and soft atmospheric glow. Brightness needs to favor the eyes, cheekbones, reactor edge, and shoulder mantle so the face remains readable at small UI sizes. The composition should be clean, centered, and compact, leaving enough breathing room around the silhouette for future card-frame use.
Subject:
Show the exact same heroine from the reference model with no face drift: an original playable heroine with warm umber skin, an angular heart-shaped face, high cheekbones, a slim slash scar through the left eyebrow, alert amber-brown eyes, and short ivory-white hair cut into an undercut with one long braided tail. Preserve the undercut, braided tail origin, eyebrow scar, cheekbone structure, nose bridge, lip shape, neck length, and athletic upper-body read. Keep only the upper costume zones visible, especially the teal mantle, ember-orange underlayer, crescent reactor, and graphite collar hardware. The expression should be calm, alert, slightly challenging, and suitable for a player selection screen rather than a cinematic battle scream.
Important details:
This image must optimize identity recognition at smaller scale. Eye direction, facial landmarks, palette blocking, and edge lighting should all support instant recognition. Avoid dramatic depth blur that would soften the face. Keep the background abstract and non-narrative, using restrained geometric layering instead of scenery. Skin texture should stay believable, hair fibers should remain stylized but not waxy, and costume material separation must still be clear even in the tighter crop.
Use case:
Hero Reveal Turnaround Pack series study 3/9, building a locked-identity bust crop for select-card and profile surfaces.
Constraints:
No readable text, no watermark, no logos, preserve exact 1:1 aspect ratio, preserve exact face and hair base from the reference model, no extreme lens distortion, no screaming expression, no cropped-off chin, no random UI text blocks.使用方法
- 复制完整原始提示词,并保留其中的参数。
- 按需求替换主体、文字和构图要求。
- 在兼容的外部工具中使用,核对当前使用条件并检查结果。
开始前需了解
- 先确认是否需要参考图。标注兼容模型,不代表该示例由这个模型生成。
- 核对名称、数字与画面文字;不同模型版本的结果可能不同。





