Thư viện prompt AI OpenZhenZhen · gpt-image-2.5 · image

Bảng tham chiếu nhân vật đa khung phong cách điện ảnh

Cinematic Multi-Panel Character Reference Board

Bảng tham chiếu nhân vật đa khung phong cách điện ảnh — khám phá tác phẩm được tạo bằng gpt-image-2.5 trên OpenZhenZhen, kèm prompt gốc và ví dụ tạo nội dung để tham khảo.

Bảng tham chiếu nhân vật đa khung phong cách điện ảnh — AI image creation
Bảng tham chiếu nhân vật đa khung phong cách điện ảnh

Prompt

Prompt tiếng Trung

使用绑定到identity_reference（ref-10a-v1）的唯一整幅照片，创作一张高级写实横向人物角色开发参考板。该整张照片既是内容的唯一来源，也是人物身份的唯一参考，不使用示例网址代替。照片中是一位成年女性的居中上半身肖像：棕色中分长发、直视镜头的平静目光、黑色细肩带上衣，背景为纯净青蓝色。整张板上的所有面板都须一致保留她可辨认的身份及所有可见特征。不得换成相似面孔，也不得重新设计她的面部、头发、可见服装或身体。 画布与版面 制作一张宽横向人物表，优先采用16:9；使用无缝中性灰摄影棚背景，以及干净、纤细、中性的面板分隔线。版面应像专业电影、时尚或游戏角色参考板，不要像随机拼贴。采用平衡柔和的摄影棚布光、真实接触阴影、自然透视与景深、栩栩如生的皮肤和材质细节、一致的颜色与光线，并保持面部清晰锐利。使用当前图像工具实际支持的最高输出分辨率，不要虚构具体分辨率保证。不要添加标志、水印、装饰元素或不必要文字。 A——三个全身视图 三个视图占据最大的组合区域：自然正面、直接背向镜头的背面，以及一个约45–90度的四分之三/侧面视图。每个视图都从头到脚完整入镜，解剖结构、姿势尺度、发型、服装和光线保持一致。保留照片可见的面部、成年年龄感、比例、肩膀、手臂、躯干、棕色长发和黑色细肩带上衣。原照片在躯干以下被裁切，且没有拍到鞋。仅对新展现的下半身作保守且视觉连贯的延伸：可采用朴素的深炭灰色下装和简洁无装饰的深色鞋；这些只是推断占位，并非已核实的原服装或事实。不得把它们说成已确认信息，不得添加珠宝或配饰，也不得改动原照可见服装。 B——三至四张面部研究 加入3–4张同一女性的面部特写：正面、四分之三角度、侧面；若为四张，再加一张强调眼睛与自然皮肤纹理的局部特写。每张都保持相同身份、脸型、眼部外观、发际线、中分、发色、可见发型和成年年龄感。采用真实摄影细节，不要美颜滤镜式磨皮。 C——两至三张表情/半身研究 加入2–3张较小的电影感半身研究：中性自信、浅笑或沉思；可选再加一种克制自然的表情或姿势。保持身份、发型、可见黑色上衣和身体比例一致；表情不得改变她的面容。手部与姿势须自然且符合解剖结构。 D——基于来源可见内容的服装细节网格 加入一组紧凑的高细节裁切图，只呈现照片中确实可见的项目：黑色上衣、细肩带、领口、可见面料表面，以及真正能看清的缝线或结构细节。不得虚构纽扣、拉链、珠宝、包、腰带、纹身、疤痕、眼镜或其他配饰。不得制作鞋履细节格，因为来源照片没有显示鞋。绝不能把隐藏的下半身或鞋履推断当成已核实的参考信息。 一致性与输出 每个面板都呈现同一位成年女性，仿佛在同一场连贯的摄影棚拍摄中完成；身份、头发、可见服装、解剖结构、比例、材质和光线相互匹配。仅把来源实际可见的特征作为事实。只有在全身视图确有需要时，才对隐藏细节作保守且视觉连贯的选择，且不得暗示这些内容来自原图观察。避免换脸、改变眼色/发型/服装、肢体重复或畸形、手部扭曲、配饰不匹配、过度模糊、伪影、无关角色、标志、水印和不必要文字。只输出一张完成的横向角色参考板，包含三个全身视图、3–4张面部研究、2–3张表情/半身研究，以及仅基于来源可见内容的服装细节网格。

Prompt tiếng Anh

Create one premium, photorealistic landscape character-development sheet from the single complete photograph bound to identity_reference (ref-10a-v1). This full image is the sole source of content and the sole person-identity reference, not an example URL. It shows one adult woman in a centered upper-body portrait: long center-parted brown hair, a direct neutral gaze, and a black thin-strap top against a plain teal-blue background. Preserve her recognizable identity and every visible feature consistently across the entire sheet. Do not replace her with a lookalike or redesign her face, hair, visible clothing, or body. CANVAS AND BOARD Make one wide landscape sheet, preferably 16:9, with a seamless neutral-gray studio background and clean, thin, neutral panel divisions. Organize it as a professional film/fashion/game character reference board, not a random collage. Use balanced soft studio lighting, realistic contact shadows, natural perspective and depth, lifelike skin and material detail, consistent color and lighting, and crisp facial focus. Use the highest output resolution actually supported by the active image tool; do not invent a numeric resolution guarantee. Add no logos, watermark, decorative elements or unnecessary text. A — THREE FULL-BODY VIEWS Give the three views the largest combined area: a natural front view, a directly turned-away back view, and one 45–90-degree three-quarter/side view. Show each figure from head to feet, with consistent anatomy, pose scale, hairstyle, clothing and lighting. Preserve the visible face, adult appearance, proportions, shoulders, arms, torso, long brown hair and black thin-strap top from the photograph. The source crops the body below the torso and does not show footwear. For the newly revealed lower body only, use a conservative, visually coherent plain charcoal outfit continuation and simple unadorned dark footwear; these are inferred placeholders, not verified original garments or facts. Do not imply otherwise, add jewelry or accessories, or alter visible source clothing. B — THREE OR FOUR FACE STUDIES Include 3–4 close-up portraits of this same woman: front-facing, three-quarter, side profile, and (if four) a close detail emphasizing the eyes and natural skin texture. Keep the same facial identity, face shape, eye appearance, hairline, center part, hair color, visible hair arrangement and adult age in every view. Use realistic photographic detail, not beauty-filter smoothing. C — TWO OR THREE EXPRESSIONS / HALF-BODY STUDIES Add 2–3 smaller cinematic half-body studies: neutral and confident, a subtle smile or thoughtful look, and optionally one other restrained natural expression or pose. Keep her identity, hairstyle, visible black top and proportions consistent; expressions must not morph her face. Use realistic, anatomically correct hands and poses. D — SOURCE-GROUNDED OUTFIT DETAIL GRID Add a compact grid of high-detail crops only for items actually visible in the photograph: the black top, its thin straps and neckline, visible fabric surface, and any seam or construction detail that can genuinely be seen. Do not invent buttons, zippers, jewelry, bags, belts, tattoos, scars, glasses, or other accessories. Do not create a footwear detail tile because the source does not show shoes. Never present hidden lower-body or footwear interpretations as verified reference information. CONSISTENCY AND OUTPUT Every panel depicts the same one adult woman in one coherent studio session, with matching identity, hair, visible outfit, anatomy, proportions, materials, and light. Preserve only source-visible traits as facts. For hidden details, make conservative visually consistent choices solely where full-body views require them, and do not imply those choices were observed in the source. Avoid face swapping, changing eye color, hairstyle or outfit, duplicate or malformed limbs, distorted hands, mismatched accessories, excessive blur, artifacts, unrelated characters, logos, watermarks, and unnecessary lettering. Output exactly one finished landscape character-reference sheet with three full-body views, 3–4 facial studies, 2–3 expression/half-body studies, and the source-grounded outfit-detail grid.

Xem mô hình API

Xem thêm prompt hình ảnh và video · Tạo hình ảnh · Tạo video · Tạo trên canvas AI