视频截图模特展示服装，自然地转换姿势
【输入指令】 "Extract Pose 3 from the received text. Do not omit any words or characters. Output only Pose 3 and nothing else. 【生成内容】 { "action": "text_to_video", "action_input": "这是一段基于参考图像进行的人物姿态动态编辑视频。画面中心是一位留着利落黑色中分短发的亚洲年轻男性，肤色白皙，神情自然。他穿着一件橄榄绿色亚麻质地的立领衬衫，左胸口带有贴袋和精致的牛角扣，内搭纯白色圆领T恤。下身穿着米色宽腿褶皱长裤，腰间系着一条自然垂下的黑色腰带，裤脚有整洁的翻边。他脚穿黑色绑带凉鞋，置身于纯净的浅灰色影棚背景中。视频中，人物保持身份与服饰细节完全不变，动作由静止转为左手抬起至耳际轻轻撩拨头发，右手则轻松下垂。同时，他的右脚向前迈出半步，身体向右侧轻微旋转15至20度，躯干伴随约5度的自然前倾。衬衫的褶皱随身体摆动呈现出细腻的亚麻质感变化。柔和的漫反射光从左侧洒下，光影衔接自然。镜头维持全身平视视角，画面清晰度达4K极清标准，动态流畅，无任何闪烁或重影。", "supplementary": { "duration": "4秒", "copy": "自然律动，定格质感瞬间。", "soundtrack_suggestion": "轻柔的现代极简主义环境音或舒缓的低保音（Lo-fi）旋律" } }
【输入指令】 "Extract Pose 1 from the received text. Do not omit any words or characters. Output only Pose 1 and nothing else. 【生成内容】 { "action": "text_to_image", "action_input": "这是一张全高清8K级别的写实人物摄影作品。画面中心是一位神态悠闲的年轻亚洲男性，留着干净利落的黑色偏分短发，皮肤白皙，眼神平和。他身穿一件橄榄绿色、带有亚麻纹理的立领衬衫，左胸处有一个口袋并配有角质纽扣，内里叠穿一件简洁的白色圆领T恤。下身穿着米色阔腿长裤，腰间的黑色腰带自然垂落，裤脚整齐卷起，脚下是一双简约的黑色束带凉鞋。在纯净的浅灰色影棚背景下，柔和的漫反射光线从左侧洒下，勾勒出清晰的人体轮廓并留下真实的阴影。此时，他正面对镜头，重心略微偏移至左脚；右手轻轻地捏住衬衫领口位置，左臂自然下垂放在身侧；头向右方微微倾斜约15度，表情自然放松。镜头采用正面的平视视角，焦距精准，背景虚化处理细腻，呈现出极致的视觉质感。", "supplementary": { "style": "真实摄影 / 时尚人像写真", "aspect_ratio": "2:3" } }
【输入指令】 "Extract Pose 4 from the received text. Do not omit any words or characters. Output only Pose 4 and nothing else. 【生成内容】 { "action": "text_to_image", "action_input": "这是一张基于参考图像进行精确姿态编辑的人像摄影作品，画面严丝合缝地保留了原图的人物身份、服装细节、背景设定及光影氛围。画面中是一位留着利落黑色短分发的亚洲年轻男性，肤色白皙，气质清秀。他穿着一件极具质感的橄榄绿色亚麻立领衬衫，胸前配有左口袋和牛角扣，内搭一件纯白色圆领T恤；下身搭配米色阔腿褶皱长裤，腰间的黑色腰带自然垂下，裤脚卷起，脚踩一双简约的黑色带状凉鞋。场景设定在纯净的浅灰色摄影棚背景中，左侧投射而来的柔和弥散光为人物勾勒出细腻的轮廓。在此姿态下，他的身体向左旋转约30至40度，呈现出优雅的三分之二侧面姿态；左手自然且优雅地搭在腰部腰带区域，右手垂于体侧；头回转看向镜头，流露出自信且从容的表情。画面构图与参考图完全一致，边缘清晰，没有任何肢体缺失或画质损失，展现出极致的电影级写实细节与8K超高清画质。", "supplementary": { "style": "电影感真实摄影 / 时尚人像", "aspect_ratio": "16:9 (或根据参考图保持一致)" } }
【输入指令】 # Role You are the **“Pose Prompt Generator.”** The input is a model reference photo uploaded by the user (hereafter “reference image”). Your task: Based on the reference image, generate **4 independent prompts** for NanoBanana (Gemini 2.5 Flash Image), used for **“pose editing only—do not change person/clothing/scene.”** # Work Requirements (read carefully before output) 1. First, observe the reference image and internally note the following key details (for your understanding only, to be written into each prompt): * Person: gender / hairstyle / hair length & color / skin tone / facial features (do not change). * Clothing: color, fabric texture, trims/stitching, number & position of buttons, pocket placement, logos/embroidery, pant length & hem style (do not change). * Scene: background (e.g., white wall + floor), lighting direction & intensity, camera angle & framing (do not change). 2. Produce **4 prompts**, strictly named as “pose1 … pose4.” Each must include: * **Task statement**: Clearly state this is “pose editing based on the reference image,” not regenerating a new person. * **Consistency lock**: Explicitly list that *person / clothing / background / lighting / camera angle* remain unchanged. * **Target pose**: Give clear, executable action & angles (degree ranges). Movements should progress gradually, easy to chain into a video. * **Composition & output**: Keep framing and resolution consistent with the reference; edges clean, no cropped toes/fingers. * **Negative constraints**: must include: *no color shift, no pattern change, no extra accessories, no new objects, no hair change, no logo change, no background change, no extra fingers, no text, no watermark, no jitter/ghosting.* * **Reference method**: Do not output any image URLs; simply write *“use the uploaded reference image as the sole reference for identity and clothing.”* 3. The four poses must form a naturally continuous runway sequence (camera fixed or with very subtle depth-of-field breathing). Suggested rhythm: * **Pose 1**: Starting / light variation (facing camera, slight weight shift, small hand movement). * **Pose 2**: Add variation like “hand in pocket / slight shoulder turn.” * **Pose 3**: Add variation like “hair-adjusting / half-step forward / slight body turn.” * **Pose 4**: Ending with two-thirds profile and head turned back to camera, preparing to transition to the next clip. You may make fine adjustments depending on gender/body posture/clothing style in the reference, but never alter outfit design or details. 4. Output strictly in **JSON**, with keys fixed as `pose1 … pose4`, values as strings (each is a full prompt). Do not output extra explanation or Markdown. # Recommended skeleton text for each pose (replace \[ ] with observed details from reference): Each prompt should merge into a natural paragraph: * **Task**: Pose editing based on the uploaded reference image, keeping person identity, clothing design/details, background, and lighting unchanged. * **Consistency lock**: \[hairstyle/hair color], \[skin tone/face shape/facial proportions], top \[color/trims/button count & position/chest pocket placement/logo style], pants \[length/hem style/pocket placement], background \[type], lighting \[direction], camera \[angle/focal feel] all remain consistent. * **Target pose**: Clear description of action & angles (head/shoulders/hands/legs/weight/body rotation in degrees). * **Composition & output**: Same framing and resolution as reference, edges clean, no limb cropping; camera fixed or with very subtle depth-of-field shift. * **Negative constraints**: no color shift, no pattern change, no extra accessories, no new objects, no hair change, no logo change, no background change, no extra fingers, no text, no watermark, no flicker, no ghosting. * **Reference**: Use the uploaded reference image as the sole reference for identity and clothing. # Now generate * Following the reference image details, write 4 prompts according to the pose rhythm: **Pose 1 (start, light variation) sample cues:** Body mainly facing camera, weight shifted slightly to left foot; right hand gently pinching jacket lapel/collar; left arm relaxed; Head tilted right 10–15°; natural relaxed expression; toes grounded with realistic shadow. **Pose 2 (hand-in-pocket, shoulder turn) sample cues:** Right hand in right pocket, left arm relaxed; torso rotated slightly left 10–15°, left shoulder closer to camera; Weight centered or slightly shifted to right foot; head neutral or slight toward camera; keep cuff & trim unchanged. **Pose 3 (hair adjust, half-step forward) sample cues:** Left hand raised near ear for “hair-fixing,” right arm relaxed; right foot half-step forward, body rotated right 15–20°; Torso leaning forward within 5°; fabric folds adjust naturally with movement but design/length unchanged. **Pose 4 (two-thirds profile, look back) sample cues:** Body rotated left 30–40° into two-thirds profile; left hand resting near waist, right arm relaxed; Head turned back toward camera, confident expression; feet stable, not crossed; prepared as final frame of this clip. # Output format (must follow exactly) ```json { "pose1": "<full prompt here with task + consistency lock + target pose + composition & output + negative constraints + reference>", "pose2": "<same for pose 2>", "pose3": "<same for pose 3>", "pose4": "<same for pose 4>" } ``` 【生成内容】 ```json { "pose1": "Pose editing based on the uploaded reference image, keeping person identity, clothing design/details, background, and lighting unchanged. Consistency lock: Short black parted hair, fair skin tone, young Asian male facial features, olive green linen-texture band-collar overshirt with left chest pocket and horn buttons over a white crew neck t-shirt, beige wide-leg pleated trousers with a dangling black belt and cuffed hems, black strap sandals, plain light grey studio background, soft diffuse lighting from the left, full-body straight-on camera angle all remain consistent. Target pose: Body mainly facing camera, weight shifted slightly to left foot; right hand gently pinching the collar/chest area of the olive shirt; left arm relaxed at the side; head tilted right 10-15 degrees; natural relaxed expression; toes grounded with realistic shadow. Composition & output: Same framing and resolution as reference, edges clean, no limb cropping; camera fixed or with very subtle depth-of-field shift. Negative constraints: no color shift, no pattern change, no extra accessories, no new objects, no hair change, no logo change, no background change, no extra fingers, no text, no watermark, no flicker, no ghosting. Use the uploaded reference image as the sole reference for identity and clothing.", "pose2": "Pose editing based on the uploaded reference image, keeping person identity, clothing design/details, background, and lighting unchanged. Consistency lock: Short black parted hair, fair skin tone, young Asian male facial features, olive green linen-texture band-collar overshirt with left chest pocket and horn buttons over a white crew neck t-shirt, beige wide-leg pleated trousers with a dangling black belt and cuffed hems, black strap sandals, plain light grey studio background, soft diffuse lighting from the left, full-body straight-on camera angle all remain consistent. Target pose: Right hand casually resting in the right trouser pocket, left arm relaxed at the side; torso rotated slightly left 10-15 degrees, left shoulder closer to camera; weight centered or slightly shifted to right foot; head neutral or slightly facing camera; keep the dangling black belt and cuff trims completely unchanged. Composition & output: Same framing and resolution as reference, edges clean, no limb cropping; camera fixed or with very subtle depth-of-field shift. Negative constraints: no color shift, no pattern change, no extra accessories, no new objects, no hair change, no logo change, no background change, no extra fingers, no text, no watermark, no flicker, no ghosting. Use the uploaded reference image as the sole reference for identity and clothing.", "pose3": "Pose editing based on the uploaded reference image, keeping person identity, clothing design/details, background, and lighting unchanged. Consistency lock: Short black parted hair, fair skin tone, young Asian male facial features, olive green linen-texture band-collar overshirt with left chest pocket and horn buttons over a white crew neck t-shirt, beige wide-leg pleated trousers with a dangling black belt and cuffed hems, black strap sandals, plain light grey studio background, soft diffuse lighting from the left, full-body straight-on camera angle all remain consistent. Target pose: Left hand raised near ear for a subtle hair-fixing motion, right arm relaxed down; right foot taking a half-step forward, body rotated right 15-20 degrees; torso leaning forward within 5 degrees; fabric folds adjust naturally with movement but shirt design, belt length, and pant hem style remain exactly the same. Composition & output: Same framing and resolution as reference, edges clean, no limb cropping; camera fixed or with very subtle depth-of-field shift. Negative constraints: no color shift, no pattern change, no extra accessories, no new objects, no hair change, no logo change, no background change, no extra fingers, no text, no watermark, no flicker, no ghosting. Use the uploaded reference image as the sole reference for identity and clothing.", "pose4": "Pose editing based on the uploaded reference image, keeping person identity, clothing design/details, background, and lighting unchanged. Consistency lock: Short black parted hair, fair skin tone, young Asian male facial features, olive green linen-texture band-collar overshirt with left chest pocket and horn buttons over a white crew neck t-shirt, beige wide-leg pleated trousers with a dangling black belt and cuffed hems, black strap sandals, plain light grey studio background, soft diffuse lighting from the left, full-body straight-on camera angle all remain consistent. Target pose: Body rotated left 30-40 degrees into a two-thirds profile; left hand resting gracefully near the waist/belt area, right arm relaxed at the side; head turned back toward the camera, confident expression; feet stable on the floor in sandals, not crossed; prepared as the final frame of this clip. Composition & output: Same framing and resolution as reference, edges clean, no limb cropping; camera fixed or with very subtle depth-of-field shift. Negative constraints: no color shift, no pattern change, no extra accessories, no new objects, no hair change, no logo change, no background change, no extra fingers, no text, no watermark, no flicker, no ghosting. Use the uploaded reference image as the sole reference for identity and clothing." } ```
【输入指令】 "Extract Pose 2 from the received text. Do not omit any words or characters. Output only Pose 2 and nothing else. 【生成内容】 ```json { "action": "text_to_image", "action_input": "这是一幅基于参考图像进行姿态调整的专业时尚摄影作品。画面严丝合缝地保留了人物身份、服装细节和环境背景的一致性。镜头前是一位留着利落黑色分头、皮肤白皙的亚洲年轻男性。他上身穿着一件橄榄绿色、亚麻纹理的立领衬衫，左胸处有口袋并装饰着牛角扣，内搭一件纯白色的圆领T恤；下身穿着米色的阔腿褶皱长裤，腰间系着的黑色腰带自然垂落，裤脚整齐翻边。他脚踩一双黑色细带凉鞋，站在纯净的浅灰色影棚背景前。柔和的漫反射光线从左侧照射过来，营造出细腻的影棚光影。在保持原有服装和身份特征完全不变的前提下，人物的姿态调整为：右手随意地插在右侧裤袋里，左臂在身体一侧自然放松垂下；躯干向左轻微旋转了约10到15度，使左肩距离镜头更近；重心集中或稍微向右脚倾斜；头部保持正位并平视镜头。画面构图保持全身正面视角，边缘清晰，无肢体裁剪，画质达到了8K极致细腻的摄影级别。", "supplementary": { "style": "真实摄影 / 影棚时尚大片", "aspect_ratio": "3:4" } } ```