
原始文档
GPT Image 2 · en
Scene:
A vertical 9:16 street-casting portrait sheet using the same 3 by 3 grid structure as the reference image: nine equal cells, thin white gutters, clean border logic, and no readable captions. The setting changes to overcast urban daylight with neutral concrete walls, subway-tile texture, glass reflections, and a casual editorial street-casting mood. The layout must stay strict even though the portraits feel more spontaneous.
Subject:
Use the reference as the primary identity guide. The same young performer appears in all nine cells with the same oval face, dark almond eyes, chin-length black hair with soft fringe, beauty mark below the left eye, and stable lip shape. Styling becomes casual streetwear without logos: black bomber jacket, white ribbed tee, gray hoodie, washed denim, plain canvas tote strap, silver ear cuff, and minimal makeup. The nine expressions are neutral casting stare, quick grin, laughing side turn, skeptical eyebrow, surprised phone-call reaction without showing a phone, soft profile, playful tongue-in-cheek smile, thoughtful downward gaze, and confident hands-in-pockets pose.
Important details:
Preserve identity from image 01 while allowing street-style variation in pose and crop. Keep each cell a clean portrait or half-body portrait, not a full street scene. The face should not age, change ethnicity, change eye color, or change hair length. The grid must remain exact, with clear gutters and no overlapping panels. Backgrounds can vary slightly but should share one urban casting-session environment and consistent overcast light.
Use case:
Nine-Expression Portrait Grid series study 4/8, testing identity lock in a casual street-casting portrait sheet.
Constraints:
No readable text, no watermark, no logos, preserve exact 9:16 vertical aspect ratio, no brand marks, no changed face identity, no extra people, no broken grid alignment, no mismatched hair, no street signs.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。





