===== A 初版 =====
Create ONE standalone horizontal 16:9 storyboard still, ideally 2048x1152, for the exact same scene and same two adult characters in the supplied references. This is a reference-preserving shot variation, not a redesign.
REFERENCE ROLES: Image 1 is the exact woman's identity, facial features, eye color, hair, hair jewelry, earrings, and costume reference. Image 2 is the exact man's identity, facial features, long hair/topknot/pin, and dark patterned costume reference. Image 3 supplies supporting continuity of these same characters. IMAGE 4 IS THE PRIMARY SCENE, SPATIAL, LIGHTING, EXPOSURE, WHITE BALANCE, BACKGROUND, AND COLOR REFERENCE. Reframe image 4 as specified below.
CRITICAL INVARIANTS: Match image 4's existing color response and light exactly: the same skin hue, neutral cream fabric, deep restrained red trim, dark brown charcoal male robe, beige/taupe dune colors, pale sky, existing shadows/highlights, contrast and saturation. No additional warmth, no new golden/orange/amber wash, no sunset enhancement, no warm filter, no teal/orange grade, no extra rimlight, no glow, bloom, haze, lens flare or relighting. Equally do not cool/desaturate/regrade the reference. Keep the existing visible source light as it is. Preserve exact character identities and natural facial proportions; no beautification or change of apparent age, hairline, makeup, skin texture, costume motifs, jewelry arrangement, headwear volume. Same sandy terrain and distant low ridges, same time and geography; do not import the blurred buildings from portraits 1/2, add new buildings, clouds, foliage, people or props. Preserve the image-4 visual/rendering style. Woman remains screen left of the man, man screen right and slightly behind her; same close interpersonal spacing, same side of the scene axis. No hugging, touching, hand actions, kiss, or extra gestures. This still is a starting frame before the line is spoken: mouth gently closed, no captions, text, watermark, borders, collage or panels.
SHOT A: A tighter close-up of the woman from the primary image 4, from upper chest to just above her original hair ornaments. Her face is left-of-center to center and occupies most of the image height. Preserve her exact upward head angle and eyes looking upward as in image 4, NOT toward the camera or the man. Her emotion is restrained sadness, only slight natural moisture along the lower eyelids, no running tears. The man's unchanged dark shoulder may remain as a small cropped presence at the far right edge, never dominate and no invented facial fragments. Same image-4 background geometry remains softly out of focus at the margins. This is a cut-in to her speech, obtained through camera framing rather than rearranging subjects. Keep full chin and enough forehead/hair so facial proportions stay natural.

===== A 连续性修订 =====
Perform a precise continuity repair to IMAGE 1, returning ONE standalone landscape still at the same dimensions/aspect ratio.
Image 1 = the current supplemental shot to refine. Image 2 = the ORIGINAL USER STORYBOARD and the authoritative reference for scene, appearance in this pose, spatial structure, LIGHTING AND COLOR. Image 3 = additional original female identity reference. Image 4 = additional original male identity reference.
Keep the close-up framing of image 1 exactly. Repair the woman's continuity details against image 2: preserve the source face's eye shape and spacing, nose width, lip contour and jaw shape rather than introducing a prettier variant; restore the original source hairline, pearl cluster and flower hair ornament placement/shape, earring construction, and visible red embroidered collar borders. Make only these reference-fidelity repairs. Her upward gaze and subtle closed-mouth emotion must remain exactly the image-2 state. Retain the small male shoulder at far right.
NON-NEGOTIABLE PRESERVATION: Match IMAGE 2's existing white balance, exposure, contrast, skin tone, fabric color, dune colors and sky. Do not increase warmth or orange/red saturation, add golden light, add rimlight, apply any cinematic filter, bloom, glow or haze. Do not cool the image either. Keep the original lighting direction and shadow softness. No extra facial shine or skin blur. Background landmark layout should correspond to a simple crop/reframe of the same IMAGE 2 terrain; do not redesign the location. Keep original visual rendering style. Keep woman on screen left and man on screen right, same scene axis. No new pose, eye contact, tears running down cheeks, hug, kiss, added objects, text, captions, panels or watermark. This is a minimal correction for consistency, not a new creative variation.

===== A 局部细节修订 =====
Edit IMAGE 1 in ONE narrowly localized area only: the red embroidered clothing borders visible below the woman's chin. IMAGE 2 is the exact original clothing reference. Match the flower ornament size, spacing, density, depth/relief and red border width to IMAGE 2, accounting only for the close-up camera scale. Remove the conspicuous oversized protruding pale-gold flower appliques in this area and restore the original finer, flatter, denser gold floral pattern. Keep original collar overlap. Preserve every other part of image 1 unchanged: face, eyes, nose, lips, jaw, gaze, hair, hair jewelry, earrings, pose, male shoulder, background, framing, exposure, white balance and ALL colors. Do not retouch the face. Do not add lighting, warmth, haze, bloom or filtering. Return one standalone landscape image with the same dimensions and no text.

===== B 初版 =====
Create ONE standalone horizontal 16:9 storyboard still, ideally 2048x1152, for the exact same scene and same two adult characters in the supplied references. This is a reference-preserving shot variation, not a redesign.
REFERENCE ROLES: Image 1 is the exact woman's identity, facial features, eye color, hair, hair jewelry, earrings, and costume reference. Image 2 is the exact man's identity, facial features, long hair/topknot/pin, and dark patterned costume reference. Image 3 supplies supporting continuity of these same characters. IMAGE 4 IS THE PRIMARY SCENE, SPATIAL, LIGHTING, EXPOSURE, WHITE BALANCE, BACKGROUND, AND COLOR REFERENCE. Reframe image 4 as specified below.
CRITICAL INVARIANTS: Match image 4's existing color response and light exactly: the same skin hue, neutral cream fabric, deep restrained red trim, dark brown charcoal male robe, beige/taupe dune colors, pale sky, existing shadows/highlights, contrast and saturation. No additional warmth, no new golden/orange/amber wash, no sunset enhancement, no warm filter, no teal/orange grade, no extra rimlight, no glow, bloom, haze, lens flare or relighting. Equally do not cool/desaturate/regrade the reference. Keep the existing visible source light as it is. Preserve exact character identities and natural facial proportions; no beautification or change of apparent age, hairline, makeup, skin texture, costume motifs, jewelry arrangement, headwear volume. Same sandy terrain and distant low ridges, same time and geography; do not import the blurred buildings from portraits 1/2, add new buildings, clouds, foliage, people or props. Preserve the image-4 visual/rendering style. Woman remains screen left of the man, man screen right and slightly behind her; same close interpersonal spacing, same side of the scene axis. No hugging, touching, hand actions, kiss, or extra gestures. This still is a starting frame before the line is spoken: mouth gently closed, no captions, text, watermark, borders, collage or panels.
SHOT B: A modestly wider two-person medium close shot of image 4, showing both actors from upper waist/chest to above their heads. Do not turn this into a distant wide or full-body view. The woman remains on the left foreground and the man on the right slightly behind her, preserving their exact close placement and torso orientation from image 4. Both faces and the existing hair ornaments remain clearly recognizable. The woman continues the exact upward gaze/head angle from image 4, restrained emotional expression, lips gently closed. The man's mouth is closed and his eyes remain on the woman's face, listening with quiet suppressed sadness. Reveal only a little more of their existing costume and the same dune backdrop; no new gesture or new physical interaction. The framing must visibly be wider than image 4 while preserving identical facial anatomy and scene layout. Natural proportions, no enlarged heads, no inflated robe volume.

===== B 连续性修订 =====
Perform a precise continuity repair to IMAGE 1, returning ONE standalone landscape still at the same dimensions/aspect ratio.
Image 1 = the current supplemental shot to refine. Image 2 = the ORIGINAL USER STORYBOARD and the authoritative reference for scene, appearance in this pose, spatial structure, LIGHTING AND COLOR. Image 3 = additional original female identity reference. Image 4 = additional original male identity reference.
Repair the unsupported wide framing in image 1: use a slightly wider TWO-PERSON CHEST-UP composition than image 2, but crop the bottom ABOVE BOTH WAISTS so NEITHER BELT nor lower-body design nor hands is visible. Do not invent waistbands or new lower garment features. Preserve the woman-left/man-right placement, depth and spacing from image 2, both head angles and gaze directions exactly. Keep this shot modestly wider than image 2 while showing only source-supported upper costumes. Preserve original faces, flower/pearl ornaments, pins, earrings, collar overlap and embroidery from image 2. Do not introduce new terrain ridges or foreground rock details; use the same source dunes and blur structure.
NON-NEGOTIABLE PRESERVATION: Match IMAGE 2's existing white balance, exposure, contrast, skin tone, fabric color, dune colors and sky. Do not increase warmth or orange/red saturation, add golden light, add rimlight, apply any cinematic filter, bloom, glow or haze. Do not cool the image either. Keep the original lighting direction and shadow softness. No extra facial shine or skin blur. Background landmark layout should correspond to a simple crop/reframe of the same IMAGE 2 terrain; do not redesign the location. Keep original visual rendering style. Keep woman on screen left and man on screen right, same scene axis. No new pose, eye contact, tears running down cheeks, hug, kiss, added objects, text, captions, panels or watermark. This is a minimal correction for consistency, not a new creative variation.

===== B 局部细节修订 =====
无额外局部修订

===== C 初版 =====
Create ONE standalone horizontal 16:9 storyboard still, ideally 2048x1152, for the exact same scene and same two adult characters in the supplied references. This is a reference-preserving shot variation, not a redesign.
REFERENCE ROLES: Image 1 is the exact woman's identity, facial features, eye color, hair, hair jewelry, earrings, and costume reference. Image 2 is the exact man's identity, facial features, long hair/topknot/pin, and dark patterned costume reference. Image 3 supplies supporting continuity of these same characters. IMAGE 4 IS THE PRIMARY SCENE, SPATIAL, LIGHTING, EXPOSURE, WHITE BALANCE, BACKGROUND, AND COLOR REFERENCE. Reframe image 4 as specified below.
CRITICAL INVARIANTS: Match image 4's existing color response and light exactly: the same skin hue, neutral cream fabric, deep restrained red trim, dark brown charcoal male robe, beige/taupe dune colors, pale sky, existing shadows/highlights, contrast and saturation. No additional warmth, no new golden/orange/amber wash, no sunset enhancement, no warm filter, no teal/orange grade, no extra rimlight, no glow, bloom, haze, lens flare or relighting. Equally do not cool/desaturate/regrade the reference. Keep the existing visible source light as it is. Preserve exact character identities and natural facial proportions; no beautification or change of apparent age, hairline, makeup, skin texture, costume motifs, jewelry arrangement, headwear volume. Same sandy terrain and distant low ridges, same time and geography; do not import the blurred buildings from portraits 1/2, add new buildings, clouds, foliage, people or props. Preserve the image-4 visual/rendering style. Woman remains screen left of the man, man screen right and slightly behind her; same close interpersonal spacing, same side of the scene axis. No hugging, touching, hand actions, kiss, or extra gestures. This still is a starting frame before the line is spoken: mouth gently closed, no captions, text, watermark, borders, collage or panels.
SHOT C: An intimate tighter two-face close shot framed for a later turn and eye contact, but the turn HAS NOT HAPPENED in this still. Woman on left and man on right, preserving image 4's relative depth, body placement and same-side-of-axis viewpoint. Crop roughly at upper chest/shoulders with both complete faces and room for the woman's chin/neck movement; their faces dominate more than in image 4, no extreme eye crop. THE WOMAN IS STILL LOOKING UPWARD, head and gaze matching image 4, NOT looking at the man yet, not facing him, not making eye contact yet. The man looks steadily at her face with closed lips. This starting state will allow her to finish an upward-looking phrase, then turn toward the man and only speak the final confession after their eyes meet in the later video. Do not depict multiple moments or the future turn. Keep just a faint restrained sadness, no streaming tears. Identical original exposure, white balance, light direction and texture, no added dramatic light.

===== C 连续性修订 =====
Perform a precise continuity repair to IMAGE 1, returning ONE standalone landscape still at the same dimensions/aspect ratio.
Image 1 = the current supplemental shot to refine. Image 2 = the ORIGINAL USER STORYBOARD and the authoritative reference for scene, appearance in this pose, spatial structure, LIGHTING AND COLOR. Image 3 = additional original female identity reference. Image 4 = additional original male identity reference.
Keep the tighter two-face framing of image 1. Restore the original face shapes and costume-accessory details from image 2: woman's exact eye/nose/lip/jaw anatomy, original flower and pearl cluster geometry, earring elements and red collar trim; man's exact facial proportions, original hair knot/pin and collar pattern/overlap. Do not re-beautify or smooth either face. The woman MUST continue looking UPWARD as in image 2. She has not turned and they have not made eye contact. The man still looks at her. Both complete faces stay visible with room for her future neck turn.
NON-NEGOTIABLE PRESERVATION: Match IMAGE 2's existing white balance, exposure, contrast, skin tone, fabric color, dune colors and sky. Do not increase warmth or orange/red saturation, add golden light, add rimlight, apply any cinematic filter, bloom, glow or haze. Do not cool the image either. Keep the original lighting direction and shadow softness. No extra facial shine or skin blur. Background landmark layout should correspond to a simple crop/reframe of the same IMAGE 2 terrain; do not redesign the location. Keep original visual rendering style. Keep woman on screen left and man on screen right, same scene axis. No new pose, eye contact, tears running down cheeks, hug, kiss, added objects, text, captions, panels or watermark. This is a minimal correction for consistency, not a new creative variation.

===== C 局部细节修订 =====
Edit IMAGE 1 in ONE narrowly localized area only: the man's dark robe collar/lapel borders below his neck. IMAGE 2 is the exact original clothing reference. Restore IMAGE 2's narrow, subdued, fine dark-gold woven pattern, its scale and original collar layer/overlap. Reduce the thick bright gold curled border pattern currently in image 1; it must look like the same restrained dark patterned robe in IMAGE 2, just filmed closer. Keep the pale inner collar identical to the source. Preserve EVERY other part of IMAGE 1 unchanged: both faces and their facial proportions, skin texture, hair, pins, woman's jewelry and costume, both gazes and head angles, positions, background, lighting, exposure, white balance and framing. The woman still looks upward, not at the man. Do not retouch or beautify faces, redesign ornaments, add warm light or apply any filter. Return one standalone landscape image with the same dimensions and no text.