保持角色一致性的图像编辑横幅
一张电影级写实风格的横幅，展示了同一位女性角色在原始方形插图与扩展后的城市场景中保持高度一致的效果。
gpt-image-2 · App / 网页设计 · 产品营销 · 城市风光 / 街道 · 摄影 · 电影 / 电影剧照 · 网红 / 模特
中文提示词
创作一张电影级宽幅横幅图像，展示 AI 图像编辑与一致性角色生成能力。场景设定为黄昏时分阴郁的旧城区街道，两侧高楼林立，地面湿润，有停放和行驶的车辆，柔和的路灯，并带有深色晕影。在同一个连续构图中呈现三个视觉区域：左侧为昏暗空旷的城市街道，没有主体人物，色彩饱和度低且略显模糊；中间是一个由细白边框勾勒的清晰方形插图，展示了同一位女性站在城市街道上的写实肖像；右侧是同一位女性全身出现在另一个风格匹配的街道场景中，手持外带咖啡杯。该女性为 {argument name="character name" default="一位 30 岁出头、气质时尚的女性"}：留着齐肩波浪深棕色长发，肤色中等偏白，妆容自然，表情沉稳自信，视线略微向左，身穿深炭灰色长款大衣、灰色针织毛衣、黑色高腰长裤，背着一个小巧的黑色单肩包。中间的插图应呈现为原始生成的方形图像，而左右两侧则呈现为外绘扩展以及同一角色与风格的变体。采用写实摄影风格，浅景深，柔和的夜色光影，右侧温暖的灯光与左侧冷蓝灰色的阴影形成对比，高动态范围，电影级调色，湿润鹅卵石路面与沥青路面上的真实倒影。画面中不包含任何文字、Logo、UI 控件或水印。
原始提示词
Create a cinematic wide banner image demonstrating AI image editing and consistent character generation. The scene is a moody urban street at dusk in a dense old city, with tall buildings on both sides, wet pavement, parked and moving cars, soft streetlights, and a dark vignette. Show exactly three visual zones across one continuous composition: on the left, a dim empty city street with no main person, desaturated and slightly blurred; in the center, a sharp square inset framed by a thin white border containing a realistic portrait of the same woman standing on a city street; on the right, the same woman appears again full-body in a different but matching street setting, holding a takeaway coffee cup. The woman is {argument name="character name" default="an unnamed stylish woman in her early 30s"}: shoulder-length wavy dark brown hair, medium-light skin, natural makeup, calm confident expression, looking slightly to her left, wearing a long dark charcoal overcoat, gray knit sweater, black high-waisted trousers, and a small black shoulder bag. Make the center inset look like the original generated square image, while the left and right sides look like outpainted extensions and a variation of the same character and style. Use photorealistic editorial photography, shallow depth of field, soft evening light, warm lamps on the right contrasting with cool blue-gray shadows on the left, high dynamic range, cinematic color grading, realistic reflections on wet cobblestones and asphalt. No text, no logos, no UI controls, no watermark.