原帖的 prompt 和手法都写在 ALT 里，我照着做了一组：同一个低角度仰拍，四个距离。 Prompt: photorealistic, rear low-angle view looking up escalator, one young Japanese woman from behind walking up mall escalator, long wavy blonde hair to mid-back, fair skin, slim legs, black off-shoulder long-sleeve top, ultra-short distressed light-blue denim shorts with frayed hems and belt, black platform heels, large brown monogram tote on shoulder, white paper bag in hand, black escalator steps with yellow lines, glass metal railings, bright mall lights, partial luxury signage background, natural skin and fabric texture, candid photo, 35mm, sharp, her back turned squarely to the camera. CAMERA GEOMETRY: the camera sits behind her at hip-to-thigh height and tilts upward; her hips and the back pockets of her shorts occupy the upper middle of the frame while the back of her thighs runs downward; the top surface of the black escalator step with its yellow edge runs along the bottom edge of the frame, and her lower legs and feet stay out of frame below. 動作の途中、iPhoneで急いで撮った写真 CAMERA GEOMETRY（本组重点，几何锁机位）: 相机在她正后方、臀部到大腿的高度向上仰；臀部与短裤后袋落在画面中上部；画面下缘是扶梯踏板的顶面；看不到发顶、看不到鞋底、看不到脸。 （原作者手法：用这段 prompt 生成后，再追加入力「接写」；本组沿用同一 prompt 与角度。原 prompt 尾部的 no face visible 我换成了正向描述 her back turned squarely to the camera，因为本地模型对否定句会反向执行。） rawsgirl · 1050×1400 · seeds 318472 / 654901 / 927365 / 142803