复制
[1] 视频生成|MiniMax-Hailuo-H3|mixed2video|ratio=16:9 resolution=2K duration=15|视频节点 3
subject_definitions:
: The same young adult Chinese male from < {{Mixed 2}} >. Preserve his exact hand-drawn 2D face, brown eyes, pale-blond hair, black high-collar Chinese-inspired outfit, layered lower garment, wide trousers, platform shoes, silver chains, circular black pendant, anatomy, ink lines and cel shading.
< {{Mixed 1}} >: The exact final 16:9 frame from Clip 1: two matte-black torn-paper fields, one jagged ivory diagonal seam from lower left to upper right, and a moving vermilion brush tip on the seam. This is the exact splice frame at 0.00s.
: A 2D motion and editing reference only. Transfer its torn-paper action transitions, diagonal manga framing, flat brush arcs, 2D smear frames, silhouette relays, extreme crop changes and rapid poster rhythm. Do not transfer its character, gender, face, outfit, weapon, text, logo, story or exact layouts.
summary:
[keyframe completion + reference generation + video continuation] A seamless 15-second 16:9 pure-2D continuation. Begin exactly from , continue its paper and brush motion, then escalate through dynamic manga crops, graphic action, silhouette relays and exact typography before an active illustrated hero poster. is the only character; controls only 2D motion grammar and edit energy.
retention_analysis:
: fully_preserved — retain the same 2D face, hair, outfit, pendant, chains, anatomy, line art and cel shading.
< {{Mixed 2}} >: attribute_transfer — retain identity and wardrobe only.
< {{Mixed 1}} >: fully_preserved — reproduce its opening paper shapes, seam, brush position and movement at 0.00s.
: partially_preserved — transfer only 2D action packaging, cut rhythm and transition behavior.
detailed_description:
Every frame remains visibly hand-drawn 2D cel animation with ink contours, flat colors, hard two-step shadows, screen-print grain, dry-brush texture, paper fibers and occasional two-frame smears. No photoreal human, live-action image, realistic skin, 3D character, CGI surface, volumetric lighting or photographic lens look.
The only readable text is “江南第一深情”. Palette: parchment ivory, ink black, vermilion red, tiny cyan registration accents. Maintain one character and no weapon. Continue cuts through pose, diagonal direction, brush motion and overlapping sound. Do not fuse hard cuts into a continuous morph.
[Shot 1] 0.00–0.40s. Begin exactly from . The red brush tip continues upward along the jagged ivory seam while both black paper fields keep closing.
[Shot 2] 0.40–0.95s. The seam tears open explosively and reveals in a wide 2D action stance on an ivory field, red brush arc behind him.
[Shot 3] 0.95–1.55s. Hard cut to the platform shoe striking a black ink line. Dry-brush cracks race diagonally upward.
[Shot 4] 1.55–2.20s. Match the diagonal into a waist-and-chain crop. The attached chain whips once while a flat red belt-like graphic crosses behind, not replacing the real outfit.
[Shot 5] 2.20–2.85s. Torn-paper window on black reveals his clean 2D profile. Hair strands and eye stay faithful to .
[Shot 6] 2.85–3.55s. Full-body figure inside an off-center black rectangle. He pivots once; two red-and-black 2D smear silhouettes record the direction, then vanish.
[Shot 7] 3.55–4.25s. Three diagonal manga panels show eye, collar and stepping foot. The panels slide in one shared lower-left to upper-right vector.
[Shot 8] 4.25–5.00s. A wide red brush wedge sweeps across the panels and reveals a low-angle forward step. Coat layers remain anatomically and structurally stable.
[Shot 9] 5.00–5.80s. Hard cut to a black field: ivory silhouette at left, real character at right. Both perform the same pose; the silhouette is a flat graphic shadow, not a second person.
[Shot 10] 5.80–6.60s. Fast crop relay—eye, pendant, hand, full body—alternating ivory, red and black. Each crop receives one dry synchronized impact.
[Shot 11] 6.60–7.40s. Overhead poster view. A red brush circle remains visibly hand-painted and incomplete while crosses its center.
[Shot 12] 7.40–8.20s. The incomplete circle opens into a torn ivory title plane. All movement clears away from the text-safe center.
[Shot 13] 8.20–9.45s. Exact title card. Show “江南第一深情” as one intact horizontal line in large flat ink-black Chinese type on parchment ivory. Place small at the far right with a red offset silhouette behind him. No glow, bevel, gold, random text or overlapping graphics. All six characters are correct, unobstructed and fully readable for at least 0.85s.
[Shot 14] 9.45–10.20s. After clean exposure, the whole title plane tears away as one sheet. Do not distort or repeat its characters.
[Shot 15] 10.20–11.00s. Reveal a fast lateral stride with one long hand-drawn coat smear and a red brush slash continuing the same direction.
[Shot 16] 11.00–11.80s. Four microcuts: shoe push, chain recoil, profile, direct eye. Use extreme scale contrast and alternate black/ivory fields.
[Shot 17] 11.80–12.65s. Low-angle full-body pose inside a tilted red frame. Black brush corners close toward him, then stop before contact.
[Shot 18] 12.65–13.45s. His silhouette becomes a flat black poster cutout; an ivory outline and red offset shadow register behind it by a few pixels.
[Shot 19] 13.45–14.20s. Paper rips vertically to reveal his exact colored face in close-up. He looks directly at camera while the tear edges continue moving.
[Shot 20] 14.20–15.00s. Final 2D hero poster: slightly right of center, low-angle three-quarter pose, huge vermilion dry-brush arc behind him, black torn-paper corners and sparse cyan flecks. Hair, coat hem, paper grain and brush edge remain subtly active; no freeze and no additional text.
overall_soundscape:
Continue Clip 1's paper drag and brush rise. Add hard paper tears, dry brush swipes, ink scratches, cloth snaps, steps and short graphic impacts. Let the next sound begin before each cut and the previous tail cross it. Keep paper movement active at the end.
non_diegetic_music:
Continue the same 164 BPM graphic breakbeat without restarting: identical kick, snare, elastic sub, paper percussion and distorted synth stabs. Drop briefly before the title, then return with denser subdivisions. End active and unresolved.
[2] 视频生成|MiniMax-Hailuo-H3|text2video|ratio=16:9 resolution=2K duration=15|视频节点 2 - 副本
subject_definitions:
: The young adult Chinese male from < >. Preserve his exact 2D illustrated face, brown eyes, tousled pale-blond hair, black high-collar Chinese-inspired jacket, layered lower garment, wide trousers, platform shoes, silver chains, circular black pendant, anatomy, line art and cel shading. He remains a hand-drawn 2D anime character in every frame.
: A visual-style and editing-rhythm reference only. Transfer its flat 2D graphic-animation grammar, cream/black/red palette, brush texture, torn-paper transitions, manga crops, scale jumps and rapid editorial cadence. Do not transfer its character, gender, face, hair, clothing, weapon, text, logo, story or exact layouts.
summary:
[reference generation] A 15-second 16:9 premium 2D graphic-anime character opener. supplies the only character identity; supplies only 2D visual grammar and edit rhythm. Use parchment ivory, ink black, vermilion red and tiny cyan registration accents. Dominant chain: graphic ignition -> identity fragments -> outfit proof -> exact name card -> torn-paper splice bridge. Deliver at least fourteen distinct editorial blocks with hard cuts and flowing screen direction.
retention_analysis:
: fully_preserved — keep the same 2D face, hair, outfit, pendant, chains, proportions, line art and cel shading.
< >: attribute_transfer — transfer character identity and wardrobe; ignore its four-panel layout and sunset background.
: partially_preserved — transfer only 2D design language, motion density, transition grammar and pacing.
detailed_description:
Every frame is visibly hand-drawn 2D cel animation: clean ink contours, flat color fills, two-step shadow shapes, screen-print grain, dry-brush edges and paper texture. Keep facial features graphic and illustrated. No live-action human, photoreal skin, realistic photography, 3D character, CGI material, volumetric lighting or cinematic lens rendering.
The only readable text is “童锦程”. Maintain one character. Use one vermilion brush stroke as the recurring carrier; it may become an underline, crop edge, action arc or torn-paper guide. Hard cuts remain fluid by inheriting pose, gaze, brush direction or sound. Do not fuse the sequence into slow morphing. No shot exceeds one second except the name card.
[Shot 1] 0.00–0.45s. Parchment-ivory field. appears as a small full-body 2D figure left of center. A vermilion brush stroke attacks horizontally behind him with a dry paint snap.
[Shot 2] 0.45–1.05s. Hard cut closer. Oversized abstract black blocks slam behind his silhouette while the red stroke continues right; the blocks never form readable text.
[Shot 3] 1.05–1.65s. Extreme 2D eye close-up. The red stroke crosses beneath the iris as a flat graphic mask; one two-frame white flash reverses the palette.
[Shot 4] 1.65–2.30s. Match cut to his platform shoe landing on a black ink baseline. The line fractures into dry-brush splinters moving right.
[Shot 5] 2.30–3.00s. Diagonal three-panel composition: attached silver chain, solid black pendant, coat closure. Each object remains structurally faithful to .
[Shot 6] 3.00–3.70s. A rough ivory paper tear crosses the panels and reveals a full-body pose inside a tilted black rectangle. Coat and hair respond with restrained hand-drawn motion.
[Shot 7] 3.70–4.45s. Hard cut to a clean profile on ivory. The red stroke hooks around his shoulder and exits without becoming a physical object.
[Shot 8] 4.45–5.25s. Match the shoulder line into an extreme collar-and-jaw crop. Switch from ivory field to solid black while preserving head direction.
[Shot 9] 5.25–6.05s. Four rapid manga crops—eye, hand, pendant, shoe—strike on separate beats. One detail per crop, no duplicate complete bodies.
[Shot 10] 6.05–6.85s. His coat sweeps left to right as a flat black shape and becomes a brush-edged wipe into a high-angle full-body composition.
[Shot 11] 6.85–7.65s. Low-angle 2D hero pose against a huge vermilion rectangle. Sparse cyan registration flecks appear only at the outer corners.
[Shot 12] 7.65–8.75s. Exact name card. On clean parchment ivory, place small at the far right. Show “童锦程” left-aligned in large flat ink-black Chinese type with refined editorial spacing. No glow, outline, bevel, metal or random text. All three characters are correct, unobstructed and fully readable for at least 0.80s.
[Shot 13] 8.75–9.50s. A torn black paper strip rips through the card diagonally; after the clean exposure the type disappears as one intact layer, never deforming into misspelled text.
[Shot 14] 9.50–10.30s. The tear reveals making one fast lateral step. Use two hand-drawn smear frames and a red brush arc behind the coat.
[Shot 15] 10.30–11.10s. Three hard cuts on the same action: shoe push-off, coat hem extension, direct eye. Alternate ivory, black and red fields.
[Shot 16] 11.10–11.95s. Full-body silhouette on black; an ivory paper cutout of his exact outline slides behind it, producing one controlled 2D offset shadow.
[Shot 17] 11.95–12.80s. Close profile framed by an irregular torn-paper window. Hair and eye remain exact; the red stroke crosses the foreground and exits upper right.
[Shot 18] 12.80–13.65s. Wide poster composition: full-body centered, a large off-axis red brush crescent behind him, black dry-brush corners and generous ivory negative space.
[Shot 19] 13.65–15.00s. Splice bridge. Two matte-black torn-paper fields close from upper left and lower right, leaving one jagged ivory diagonal seam from lower left to upper right. A vermilion brush tip travels upward along the seam. Preserve the exact seam shape, brush position and motion in the final frame; never freeze.
overall_soundscape:
Dry paper rips, brush attacks, ink scrapes, cloth snaps, one grounded step and short graphic impacts. Use J-cuts and L-cut tails so every hard cut flows. End with the unresolved paper drag and rising brush sound.
non_diegetic_music:
Fast 164 BPM graphic breakbeat: dry kick, clipped snare, elastic sub, paper percussion and short distorted synth stabs. Use brief silence before the name card. End mid-phrase without cadence.
[3] 视频生成|MiniMax-Hailuo-H3|mixed2video|ratio=16:9 resolution=2K duration=15|视频节点 2
subject_definitions:
: The young adult Chinese male from < {{Mixed 1}} >. Preserve his exact 2D illustrated face, brown eyes, tousled pale-blond hair, black high-collar Chinese-inspired jacket, layered lower garment, wide trousers, platform shoes, silver chains, circular black pendant, anatomy, line art and cel shading. He remains a hand-drawn 2D anime character in every frame.
: A visual-style and editing-rhythm reference only. Transfer its flat 2D graphic-animation grammar, cream/black/red palette, brush texture, torn-paper transitions, manga crops, scale jumps and rapid editorial cadence. Do not transfer its character, gender, face, hair, clothing, weapon, text, logo, story or exact layouts.
summary:
[reference generation] A 15-second 16:9 premium 2D graphic-anime character opener. supplies the only character identity; supplies only 2D visual grammar and edit rhythm. Use parchment ivory, ink black, vermilion red and tiny cyan registration accents. Dominant chain: graphic ignition -> identity fragments -> outfit proof -> exact name card -> torn-paper splice bridge. Deliver at least fourteen distinct editorial blocks with hard cuts and flowing screen direction.
retention_analysis:
: fully_preserved — keep the same 2D face, hair, outfit, pendant, chains, proportions, line art and cel shading.
< {{Mixed 1}} >: attribute_transfer — transfer character identity and wardrobe; ignore its four-panel layout and sunset background.
: partially_preserved — transfer only 2D design language, motion density, transition grammar and pacing.
detailed_description:
Every frame is visibly hand-drawn 2D cel animation: clean ink contours, flat color fills, two-step shadow shapes, screen-print grain, dry-brush edges and paper texture. Keep facial features graphic and illustrated. No live-action human, photoreal skin, realistic photography, 3D character, CGI material, volumetric lighting or cinematic lens rendering.
The only readable text is “童锦程”. Maintain one character. Use one vermilion brush stroke as the recurring carrier; it may become an underline, crop edge, action arc or torn-paper guide. Hard cuts remain fluid by inheriting pose, gaze, brush direction or sound. Do not fuse the sequence into slow morphing. No shot exceeds one second except the name card.
[Shot 1] 0.00–0.45s. Parchment-ivory field. appears as a small full-body 2D figure left of center. A vermilion brush stroke attacks horizontally behind him with a dry paint snap.
[Shot 2] 0.45–1.05s. Hard cut closer. Oversized abstract black blocks slam behind his silhouette while the red stroke continues right; the blocks never form readable text.
[Shot 3] 1.05–1.65s. Extreme 2D eye close-up. The red stroke crosses beneath the iris as a flat graphic mask; one two-frame white flash reverses the palette.
[Shot 4] 1.65–2.30s. Match cut to his platform shoe landing on a black ink baseline. The line fractures into dry-brush splinters moving right.
[Shot 5] 2.30–3.00s. Diagonal three-panel composition: attached silver chain, solid black pendant, coat closure. Each object remains structurally faithful to .
[Shot 6] 3.00–3.70s. A rough ivory paper tear crosses the panels and reveals a full-body pose inside a tilted black rectangle. Coat and hair respond with restrained hand-drawn motion.
[Shot 7] 3.70–4.45s. Hard cut to a clean profile on ivory. The red stroke hooks around his shoulder and exits without becoming a physical object.
[Shot 8] 4.45–5.25s. Match the shoulder line into an extreme collar-and-jaw crop. Switch from ivory field to solid black while preserving head direction.
[Shot 9] 5.25–6.05s. Four rapid manga crops—eye, hand, pendant, shoe—strike on separate beats. One detail per crop, no duplicate complete bodies.
[Shot 10] 6.05–6.85s. His coat sweeps left to right as a flat black shape and becomes a brush-edged wipe into a high-angle full-body composition.
[Shot 11] 6.85–7.65s. Low-angle 2D hero pose against a huge vermilion rectangle. Sparse cyan registration flecks appear only at the outer corners.
[Shot 12] 7.65–8.75s. Exact name card. On clean parchment ivory, place small at the far right. Show “童锦程” left-aligned in large flat ink-black Chinese type with refined editorial spacing. No glow, outline, bevel, metal or random text. All three characters are correct, unobstructed and fully readable for at least 0.80s.
[Shot 13] 8.75–9.50s. A torn black paper strip rips through the card diagonally; after the clean exposure the type disappears as one intact layer, never deforming into misspelled text.
[Shot 14] 9.50–10.30s. The tear reveals making one fast lateral step. Use two hand-drawn smear frames and a red brush arc behind the coat.
[Shot 15] 10.30–11.10s. Three hard cuts on the same action: shoe push-off, coat hem extension, direct eye. Alternate ivory, black and red fields.
[Shot 16] 11.10–11.95s. Full-body silhouette on black; an ivory paper cutout of his exact outline slides behind it, producing one controlled 2D offset shadow.
[Shot 17] 11.95–12.80s. Close profile framed by an irregular torn-paper window. Hair and eye remain exact; the red stroke crosses the foreground and exits upper right.
[Shot 18] 12.80–13.65s. Wide poster composition: full-body centered, a large off-axis red brush crescent behind him, black dry-brush corners and generous ivory negative space.
[Shot 19] 13.65–15.00s. Splice bridge. Two matte-black torn-paper fields close from upper left and lower right, leaving one jagged ivory diagonal seam from lower left to upper right. A vermilion brush tip travels upward along the seam. Preserve the exact seam shape, brush position and motion in the final frame; never freeze.
overall_soundscape:
Dry paper rips, brush attacks, ink scrapes, cloth snaps, one grounded step and short graphic impacts. Use J-cuts and L-cut tails so every hard cut flows. End with the unresolved paper drag and rising brush sound.
non_diegetic_music:
Fast 164 BPM graphic breakbeat: dry kick, clipped snare, elastic sub, paper percussion and short distorted synth stabs. Use brief silence before the name card. End mid-phrase without cadence.
[4] 视频生成|star-video2.5|mixed2video|ratio=16:9 resolution=720p duration=30 enableSound=on|视频节点 4
【生成目标】
生成一条以真实人物表演为核心的室内夜景双人暧昧对话短片。叙事重点不是甜宠感,而是两个成年人在工作结束后的餐桌边互相试探。女主从一开始就比男主更确定、更深情,她主动把关系问题摆到台面上;男主前半段保持熟练、松弛、稍微轻浮的撩人状态,习惯用笑和反问化解问题,后半段被女主连续击中后逐渐收起职业性笑容,最终第一次认真发出私人邀约。
全片情绪弧线固定为:女主主动试探,男主轻松接招,女主持续逼近,男主第一次失去节奏,男主重新组织,男主认真邀约,女主确认并接受。
【参考素材职责】
{{Mixed 1}} 定义整段视频的唯一场景基准,包括室内空间布局、木质桌椅、中央拍摄设备、窗外夜景、植物、暖色吊灯、冷青环境光、低照度曝光、暖橙与冷青交织的综合色温、空气透视和整体电影级影调。人物进入画面后必须接受同一组环境光逻辑,不能出现人物像后贴进场景的分离感。
{{Mixed 2}} 只定义女主的五官、脸型、发型、发色、真实皮肤特征、身材比例和白色丝质上衣、白色长裤、白色高跟鞋。只生成一个女主实例,不采用参考图原背景和参考图排版。
{{Mixed 3}} 只定义男主的五官、脸型、浅金色短发、发型轮廓、身材比例和黑色中式廓形服装。只生成一个男主实例,不采用参考图原背景和参考图排版。
@音频1只用于男主实际开口时的音色、共鸣位置、自然语速和说话质感。对白文字以本提示词为唯一内容来源,不复述@音频1中的原句。女主声音不使用@音频1。
【空间与人物关系】
两人位于 {{Mixed 1}} 的木桌两侧,相对而坐,桌面中央原有拍摄设备已经停止工作,作为两人刚结束工作状态的视觉证据。两人的身体距离始终保持正常餐桌社交距离,不突然靠得过近。
摄影轴线沿两人视线建立,全部正反打和POV镜头保持同一轴线。
女主看男主时,视线落点与镜头光轴距离极小,形成直接而稳定的凝视,但不是看观众。男主看女主时同样保持准确眼线。
女主全片的身体底座是稳定、放松、稍微前倾;不通过低头、躲闪或夸张羞涩表达好感,而通过持续注视、微小嘴角变化、轻缓呼吸和有限前倾表达深情。
男主前半段的身体底座是肩背放松、身体略微后靠、下巴轻抬、单侧嘴角有很浅的笑,带一点熟练撩人感;不是油腻,不做夸张挑眉,不做喜剧式坏笑。后半段随着问题深入,身体后靠逐步消失,笑容减少,视线停留时间变长,最终进入真正认真状态。
【光线与皮肤质感】
全片严格继承 {{Mixed 1}} 的光线方向和综合色温。
暖色吊灯形成柔和暖侧光和局部高光,窗外冷青夜色提供很弱的环境填充与发丝边缘分离。人物面部亮度只比周围空间自然抬高少量,不能形成单独的美颜补光区。
暖光落在皮肤上时保留真实综合色温,冷色环境光只进入阴影和轮廓,不把皮肤整体染成橙色或青色。
所有中近景、近景、大特写、极近特写必须保持真实电影摄影皮肤:可见自然毛孔、极细面部绒毛、眼下真实纹理、轻微肤色不均、鼻翼和额头细小皮脂高光、自然唇纹、真实睫毛和眼球湿润反光。
高光有柔和滚降,阴影保留细节,焦内皮肤清楚但不锐化成数码硬边。景深浅但两只眼睛的关键焦点稳定,不出现一只眼睛清楚另一只眼睛异常失焦。
人物动作时焦点自然跟随眼睛,不出现焦点漂移。
近景与特写保持真实镜头呼吸和轻微手持漂移,每个镜头只做很小的构图修正,不做大幅推拉、环绕、升降或炫技移动。
【声音总则】
环境声只保留真实室内夜景底噪:远处极轻的人声、空调或通风系统低频、杯碟偶发轻响、窗外遥远城市声。对白始终在前景,环境声不抢字。
不新增背景音乐。
男主每次发声都使用@音频1的音色特征;女主使用自然、轻柔、成熟、克制的女声。两人都采用生活化口语节奏,不播音,不短剧腔。
每个镜头只有指定说话人形成完整口型。画面内另一人聆听时保持自然闭口,只允许自然眨眼、呼吸和极小反应,避免串口型和重复台词。
【镜头时间线】
镜头1|0.0-2.5s
景别:女主大特写,男主POV,85mm感,极浅景深。
机位:平视,女主双眼位于画面上三分线附近,背景暖灯和冷青窗景形成柔和散景。轻微手持呼吸,不主动推近。
开始状态:开场直接进入女主脸部。女主没有笑,肩颈放松,双眼稳定锁定男主,第一次自然眨眼后继续看着他。前一秒嘴角保持平直,只有眼神先建立亲密感。
表演过程:她开始说第一句时仍然不笑。说到句子中段,单侧嘴角才极轻地上扬,眼睛始终不躲。语气不是质问,而是知道答案后故意给对方机会回答。声音轻,语速偏快但咬字清楚,前半句一口气说完,随后停约半拍。
女主:{你是不是特别会让女生觉得}
停半拍。
女主:{自己是例外?}
结束状态:最后一个字落下后不追问,嘴角浅笑保留,视线继续停在男主眼睛上。
动作音效:极轻衣料摩擦,室内底噪。
镜头2|2.5-4.5s
景别:男主近景,女主POV,约65mm至85mm感。
机位:平视固定轻手持。男主上半身与部分椅背可见,背景保持@图片1的暖冷综合色温。
开始状态:男主身体略微后靠,肩膀放松,先看她半拍,单侧嘴角慢慢抬起,下巴轻抬一点。
表演过程:男主不是立刻认真回答,而是用熟练反问把问题推回去。笑意只停在嘴角,眼睛仍然观察她,带一点轻浮和玩味。音量正常偏低,语速松弛,尾音轻轻上挑。
男主使用@音频1:{那你觉得你是吗?}
结束状态:说完保持浅笑,视线不移开,等她回答。女主不在画面内。
动作音效:椅背极轻承重声,室内底噪。
镜头3|4.5-7.0s
景别:女主大特写,男主POV。
机位:延续镜头1的光轴与焦段感,构图略微更稳。
开始状态:女主仍在看男主,上一镜男主的反问成为触发。她没有出现被撩到的躲闪,只完成一次很慢的眨眼。
表演过程:她嘴角的浅笑收平,声音轻但干脆,不拖尾。
女主:{我不想当。}
女主说完立即闭口,保持视线。
男主画外使用@音频1,语气带一点好奇:{为什么?}
男主画外说话期间,女主自然闭口聆听,视线仍然固定,只在最后一个字后极轻吸气。
结束状态:女主像已经准备好下一句,眼神比开场更明确。
环境声:室内底噪保持不变。
镜头4|7.0-9.5s
景别:女主近景,男主POV,约85mm感。
机位:女主从原坐姿向前移动很小距离,相机不主动后退,人物因此在画面中自然变大一点。保持同一光轴。
开始状态:女主身体重心从椅背稍微移向桌面方向,手臂仍自然放松,目光持续锁定男主。
表演过程:她先把声音压轻一点,语速慢下来。第一句说完后嘴角短促上扬,不露齿,像是把答案说给一个她已经看透的人。
女主:{例外太多了}
停很短一拍。
女主:{就不叫例外了。}
结束状态:第二句落下时浅笑收住,只剩眼神继续停在男主脸上,形成她掌控节奏的状态。
动作音效:衣料随前倾产生极轻摩擦。
镜头5|9.5-11.5s
景别:男主近景。
机位:平视,固定轻手持,焦点牢牢落在眼睛。
开始状态:男主原本带笑,嘴唇已经准备接话。听到女主最后一句后,嘴角弧度停住,眼睛短暂地在她双眼之间重新聚焦。
表演过程:停约四分之一拍后再开口。身体不再继续后靠,下巴回到自然高度。语气仍有一点玩笑底色,但明显比镜头2认真,尾音不再上挑。
男主使用@音频1:{那你想当什么?}
结束状态:说完保持自然闭口,笑意只剩很浅一层,真正等待答案。
环境声:这一刻环境底噪保持稳定,不做戏剧性抽空。
镜头6|11.5-15.5s
景别:女主极近特写,男主POV,85mm至100mm感。
机位:平视,镜头保持静止,只保留几乎不可察觉的手持呼吸。构图以双眼、鼻梁、嘴唇为核心,保留少量发丝和面颊轮廓。暖灯在一侧眼球形成小面积高光,另一侧保留冷青环境反射。
开始状态:女主听到问题后不马上说话。她保持直视,呼吸幅度略微减小,眼神在男主双眼之间稳定停留。
表演过程:前约1.3秒保持沉默。期间只发生两个变化:下眼睑极轻收紧,嘴角原有弧度逐渐变软。她像在判断眼前这个男人有没有真正认真,而不是表现害羞。
随后声音比前面再轻半档,语速慢,字头清楚。
女主:{你不工作的时候}
停一拍,仍然直视。
女主:{还会想见的人。}
说完后不低头,不闪躲,不用笑来缓解。双唇自然闭合,眼神持续柔软地停在男主身上。
结束状态:她已经把关系问题说到最直接的位置,但身体仍然稳定,没有戏剧性动作。
环境声:轻微通风低频与远处杯碟声,保持真实。
镜头7|15.5-17.5s
景别:男主特写,约85mm感。
机位:平视固定轻手持,背景虚化但仍可辨识@图片1的暖色灯与冷青夜景。
开始状态:男主接收到女主的答案,上一镜的浅笑仍残留在嘴角。
表演过程:笑容在约半秒内自然淡掉,不突然变脸。双唇轻轻分开像准备说话,但没有形成声音,随后重新闭合。视线锁定女主,眨眼频率降低。身体前倾极小距离,像第一次从熟练应对变成真正聆听。
无对白。
结束状态:安静约一秒,只看着她,职业性笑容基本消失。
环境声:室内底噪和他一次很轻的鼻息可听见。
镜头8|17.5-19.5s
景别:女主特写,男主POV。
机位:平视,固定,保持女主双眼清晰。
开始状态:女主捕捉到男主没有接话。她不乘机夸张胜利,只让眼睑轻轻收窄一点,嘴角出现很小的笑。
表演过程:第一句轻而短,带一点温柔调侃。
女主:{怎么?}
停半拍。
女主:{突然不会说话了?}
第二句尾音微微上扬,但音量不变。眼睛始终看着男主,不做挑衅式抬下巴。
结束状态:说完保持很浅的笑,继续等他接。
环境声:远处人声极轻。
镜头9|19.5-22.0s
景别:男主近景。
机位:平视,轻微手持呼吸。
开始状态:男主先重新找回一点笑意,但这次笑意出现得比镜头2慢,且不完全回到眼睛。
表演过程:他先短促呼气,像重新把自己放回熟悉的轻松状态。第一句语速略慢,带一点自我缓冲。
男主使用@音频1:{你这个问题……}
停很短一拍。
男主使用@音频1:{不适合工作的时候问。}
第二句说完时单侧嘴角浅浅抬起,但眼神仍然认真看她,表现出嘴上在躲,注意力已经没有躲。
结束状态:男主保持自然闭口,等女主反击。
环境声:室内底噪。
镜头10|22.0-24.5s
景别:女主大特写,男主POV。
机位:平视,固定,极浅景深。焦点锁在靠镜头一侧眼睛并覆盖双眼关键区域。
开始状态:女主听到男主用工作做挡箭牌,没有马上笑。她的面部几乎静止,视线稳。
表演过程:声音非常轻,语速慢,句尾直接收住。
女主:{现在又没工作。}
说完后立刻自然闭口。
余下至少0.8秒只保留眼神和一次极轻呼气,不笑,不低头,不追加动作。双眼持续看着男主,让沉默本身成为压力。
结束状态:女主完全不退,让男主必须真正回答。
环境声:真实室内底噪完整保留。
镜头11|24.5-27.0s
景别:男主特写。
机位:平视,固定轻手持,构图比前面更克制,男主脸部占画面更大。
开始状态:男主听完后不再用明显笑容应对。他看她半拍,下颌放松,肩膀从轻微后靠回到自然中立。
表演过程:声音下降一点,语速正常,尾音平收,第一次明确像私人邀约。
男主使用@音频1:{那明天晚上。}
男主说完闭口,看着她。
女主画外,声音轻、带一点确认:{干嘛?}
女主画外说话时男主保持闭口,嘴角出现很浅的笑,不抢口型。
男主随后用@音频1,语气简单直接:{继续吃饭。}
结束状态:说完不补玩笑,不移开视线,保持认真。
环境声:极轻杯碟声和室内底噪。
镜头12|27.0-30.0s
景别:女主极近特写,男主POV。
机位:平视,85mm至100mm感,固定轻手持。双眼和嘴角是唯一表演重点,背景暖色灯形成柔和圆形散景,冷青窗景停留在阴影侧。
开始状态:女主先出现极轻的意外,表现为上眼睑抬高一点,吸气短暂停住,持续不超过半秒。
表演过程:意外很快转成柔软。眼睑重新放松,嘴角缓慢上扬但不露齿,眼神明显比开场更温柔,始终不离开男主。
女主:{这次算你约我?}
说完闭口,保持注视。
男主画外使用@音频1:{嗯。}
男主画外发声期间女主自然闭口,笑意只增加一点。
停半秒。
女主声音很轻,尾音收住,不拖长:{那我当真了。}
结束状态:最后一句说完后女主仍然直视男主,嘴角保留非常克制的含情笑意,眼神柔软且确定。停在这个状态结束,不再增加动作或台词。
环境声:室内夜景底噪自然延续到结束。
【剪辑与连续性】
12个镜头按用户给定顺序发生。每次切换都由上一句台词的落点或对方反应触发,不插入无关空镜。
女主POV与男主POV保持一致眼线和摄影轴线。正反打之间两人的坐姿高度、桌面相对位置、服装褶皱、头发状态、杯具和中央拍摄设备位置保持连续。
两人从始至终都是同一实例,不出现人物复制、服装变化或场景结构变化。
【表演收束】
女主的情绪递进只通过稳定凝视、极小嘴角变化、轻微前倾和呼吸变化完成。她的核心不是羞涩,而是深情、主动和确定。
男主的轻浮感只存在于前半段的身体后靠、浅笑、轻微下巴抬高和反问语气中。随着女主把关系说得越来越具体,这些特征逐步减少。
镜头7开始,他的视线更稳定、笑容更少、身体更中立。
镜头11的邀约必须让观众清楚感到他已经从熟练撩人切换到认真。
全片表演保持生活化和克制,不做夸张短剧反应,不用大笑、捂脸、明显害羞低头或戏剧性停顿替代细微表演。
[5] 视频生成|star-video2.5|text2video|ratio=16:9 resolution=720p duration=30 enableSound=on|视频节点 4 - 副本
【生成目标】
生成一段发生在 餐厅空间中的真人实拍电影级双人对话剧情。叙事核心不是夸张调情,而是女主播连续主动试探,童锦程从游刃有余、短暂被问住,到逐渐收起玩笑,最后暴露真实心意。表演必须生活化、克制、有停顿、有聆听反应;每次情绪变化都有明确触发,不做短剧式挤眉弄眼,不做偶像MV式摆拍。
【参考素材职责】
用于整段视频唯一场景,以及空间布局、室内陈设、桌椅位置、落地窗、植物、木质材料、桌面直播设备、夜间环境、冷暖光源关系、整体曝光、色调、影调和空气感。整段视频始终发生在这一真实空间内。
只定义女主播同一个人物实例的面部五官、发型、发量、身材比例和白色服装结构。将参考图中的棚拍光转换为@图片1真实餐厅环境光,保留人物身份特征,不采用参考图的纯色背景与棚拍构图。
只定义童锦程同一个人物实例的面部五官、浅金色短发、发量结构、黑色中式上装及整体服装轮廓。将参考图中的白色棚拍质感完全转换为 真人电影环境中的真实皮肤、真实头发和真实布料,不采用参考图背景和人物展示板构图。
@音频1只绑定童锦程本人所有对白的音色、声音质感、说话习惯和自然节奏。女主播不使用@音频1的音色。
【场景与影调】
以@图片1作为绝对视觉基准。夜间高级日式融合餐厅,室内主色保持深木棕、墨绿、暗青蓝与低饱和暖橙。顶部暖橙色吊灯是真实主光源,落地窗外的暗青蓝夜色形成自然冷色环境填充,桌面、皮革椅、植物叶片与人物皮肤均接受同一套真实光源。
人物进入近景、特写时不能突然出现独立美颜灯或棚拍补光。暖色吊灯自然落在人物额头、鼻梁、颧骨和发丝一侧,窗外冷青环境光轻微进入阴影侧,下颌和颈部仍保留真实阴影过渡。每次切换景别时人物曝光、色温、黑位和皮肤高光连续稳定,人物必须真正属于这个空间,而不是贴在背景上的独立肖像。
所有中近景、近景、面部特写保持真人电影摄影皮肤:可观察到自然毛孔、极细面部绒毛、轻微油脂高光、嘴唇纹理、眼周细微纹理、真实眼球湿润反光和自然肤色细小差异。高光柔和滚降,阴影保留细节。脸部质感不因为靠近镜头而变光滑,不出现磨皮、美颜、蜡质感、塑料感或过度锐化。头发保持大量独立发丝和真实透光,服装保持真实织物褶皱与材质响应。
两人的主空间轴线固定:女主播主要位于画面左侧,童锦程主要位于画面右侧餐厅入口与木质门框一带,两人斜向面对彼此。摄影机始终留在同一侧轴线,所有正反打保持眼神方向一致。
整段采用轻微手持观察感。手持来自摄影师呼吸和身体重量变化,只产生很轻的漂移和重新构图,不做明显抖动、甩动或MV式运动。
【声音】
保留安静夜间餐厅真实底噪、轻微空调声、远处店内人声以及落地窗外模糊城市环境声。人物对白始终清楚靠前。
背景音乐只是非常轻的环境性音乐。女主播第一句开口时音乐迅速压到近乎听不见,随后持续保持低位,让两人的呼吸、停顿和反应成为重点。0:25以后才非常轻地增加一点情绪层次,不做煽情高潮。
童锦程开口时严格使用@音频1的声音特征。每个童锦程说话阶段,女主播保持自然闭口聆听;女主播说话阶段,童锦程保持闭口,以眼神和细小表情回应,避免两人同时形成说话口型。
【镜头1|0:00-0:03】
近景,约65mm真人电影镜头感。
从女主播上半身近景直接快速进入剧情,摄影机已有极轻手持呼吸。背景仍能辨认@图片1的暖色吊灯、暗青色落地窗和模糊绿植,使人物始终处于真实空间中。
开始时女主播身体并未完全朝向童锦程。听见现场上一拍结束后,她颈部先快速转向画面右侧,头发出现极轻惯性摆动,随后身体才跟过去一点。
转到位后,她的视线立刻锁住童锦程的双眼。上眼睑轻微抬起,眉毛只出现很小的上扬,随后恢复放松。眼神不是挑衅式瞪视,而是明显带着好感和主动确认感。双唇原本自然闭合,开口前嘴角只保留极浅笑意。
她身体重心稳定,没有夸张前倾。
女主播声音自然、清楚,语速略慢,第一句直接落下,不故意撒娇:
{你是不是跟每个女生都说,下次单独约?}
最后几个字说完后,她没有马上移开视线,仍看着童锦程,给对方半拍压力。
音乐在她开口瞬间压低,餐厅真实环境底噪轻微浮出。
【镜头2|0:03-0:05】
直接正反打至童锦程中近景,约65mm至85mm,轻手持保持。
童锦程靠在画面右侧餐厅入口附近的木质门框边,背景暖色吊灯形成柔软散景,脸部一侧接受吊灯暖光,另一侧被落地窗冷青环境光轻微填充。
女主播的问题结束以后,他没有立即回答。
先把目光向下落半拍,嘴角自然上扬形成一个很浅的闭嘴笑,像是觉得这个问题有意思。随即抬眼重新找到女主播眼睛,笑意仍在,但没有大笑。
他使用@音频1音色,第一句短、轻、自然:
{不是。}
停半拍。
他的头轻微偏一点,视线仍锁住她,第二句比第一句稍亮一点,带着很熟练的玩笑感,但语气不能油腻:
{一般我不等下次。}
说完后闭口,嘴角仍保留很浅弧度,等待她的反应。
【镜头3|0:05-0:08】
近景双人偏侧构图,约50mm。女主播位于画面左侧,童锦程位于画面右侧,两人同时保持可见,中央背景仍能辨认桌面直播设备与暖灯。
摄影机只有很轻的连续前探,使两人的距离感变得更近,但不改变轴线。
女主播听完先轻轻笑一下。眼睑自然收窄,嘴角真实上扬,但笑只持续很短。随后她向童锦程方向迈半步,脚步落稳后身体才微微前倾,空间距离真实缩短。
她仍看着他的眼睛:
{那你怎么不问我?}
童锦程的眉梢快速轻抬一下,头部保持基本不动,闭口听完才接:
{问你什么?}
他说完立刻闭口。
女主播这次没有退,视线从他的双眼极短暂扫到嘴角,再重新回到眼睛,笑意收小,语速放慢一点:
{你平时怎么问,就怎么问。}
最后一个字落下后,她保持站位,不再继续靠近,把主动权完整交给童锦程。
【镜头4|0:08-0:11】
童锦程面部近景,约85mm。固定构图配合几乎不可察觉的手持呼吸,不推镜。
这一拍完全依靠真实微表情。
开始时他嘴角仍残留上一拍的浅笑。听完女主播最后一句以后,笑意没有立刻消失,但动作突然停住。
他的眼睛持续看着她,眨眼频率短暂下降。嘴角那一点弧度保持一瞬,随后极缓慢地收小。下颌没有夸张绷紧,只完成一次非常轻的吞咽。
他原本准备开口,但双唇轻微分开以后又自然合上。
整个过程中不说话。
不要让他表现成惊讶或尴尬,只表现为一个向来接话很快的人第一次真正停顿。
环境底噪在这两秒稍微变得可感,音乐继续极低。
【镜头5|0:11-0:15】
切女主播近景,约85mm,摄影机在整句期间完成一次极缓慢、极短距离的前推,最后稳定在她的眼神。
她已经看见童锦程的停顿。
开始时嘴角出现很浅的得逞笑意,一侧眉毛轻抬,眼睛仍没有离开他。她不是大笑,也不挤眉弄眼。
开口前轻轻吸一口气,身体保持稳定。
声音比第一句更轻一点,语速偏慢,前半句带一点逗他,后半句直接压向核心:
{怎么,真碰上一个敢答应的,你不会了?}
说到最后几个字时,她嘴角笑意反而减少,眼神保持稳定,让这句话从玩笑逐渐接近认真。
说完闭口,等待回答。
【镜头6|0:15-0:18】
中近景双人,约50mm至65mm。
童锦程仍在画面右侧,女主播在左侧,两人的距离延续镜头3,没有位置跳变。
童锦程先完成一次缓慢呼气。原本的玩笑笑意明显收住,下颌和肩膀都稍微松下来,不再靠表情抢节奏。
他第一次用比此前更低、更慢的声音说:
{我不是不敢问。}
说完闭口,视线不躲。
女主播几乎没有动作,只轻微抬一下下巴,声音短:
{那是什么?}
她立即闭口。
童锦程停极短一拍,视线仍停在她眼睛上。嘴角没有重新笑起来,声音更轻,句尾迅速收住:
{怕你真答应。}
说完以后不要露出得意神情。让这句话第一次产生真实分量。
【镜头7|0:18-0:21】
女主播近景,约85mm,固定轻手持。
她听见“怕你真答应”以后没有马上笑。
先完成一次自然眨眼,随后视线重新稳稳锁住童锦程。嘴角原本的浅笑稍微变软,呼吸速度略微下降,但身体完全没有后撤。
她非常轻地抿一下嘴唇后开口,声音平稳,语速慢:
{我答应,你怕什么?}
说完后双唇自然闭合。
她仍然看着他,没有害羞式躲闪,也没有夸张期待表情。情绪通过不躲开的眼神完成。
【镜头8|0:21-0:25】
先保持童锦程85mm面部近景,再使用一次独立的手机特写。
女主播的问题结束。
童锦程没有马上回答。他持续看着她大约两秒,面部几乎静止,只在吸气时鼻翼出现非常轻的变化。
随后他低声说:
{怕明天不好播。}
语气不是抖包袱,而是半认真地给自己找一个出口。
话音刚落,他手里的手机发生一次清楚、短促的震动。
镜头切至手机手部特写。真实手指、指甲和皮肤纹理清楚,手机屏幕被暖色灯光与周围环境自然反射照亮。
手机只表现为一次直播相关通知弹出和屏幕亮起,具体可读文字留给后期叠加。
女主播在画外立即问:
{为什么?}
她问完以后保持安静。
动作音效突出一次真实手机震动,随后重新回到安静餐厅底噪。
【镜头9|0:25-0:29】
双人近景,约50mm至65mm。
摄影机从稳定双人构图开始,以极慢、连续的方式向两人前方轻微靠近,最终主要停在两人的眼神关系上,不形成夸张大特写。
童锦程把手机抬到准备重新进入直播的位置。
动作做到一半,他的拇指悬在屏幕附近,没有按下。
手机停止移动。
他抬眼重新看向女主播。
这一刻嘴角已经没有之前习惯性的调笑。他的视线更稳定,声音压低半档,语速放慢:
{我要是知道你真喜欢我……}
说到这里停半拍。
停顿期间他仍看着她,不低头,不笑。
然后继续:
{我还怎么跟他们演不喜欢你?}
第二句不做情话式煽情。前半句保持克制,最后几个字声音稍微放轻,句尾自然收住。
女主播全程闭口。听到后半句以后,她的眼睑非常轻地放松,原本带压迫感的视线开始变软,嘴角没有马上笑出来。
背景音乐只在这一刻非常轻地恢复一点情绪层次,依然低于对白与环境声。
【镜头10|0:29-0:30】
直接切女主播85mm面部近景作为结尾。
她听完以后短暂愣住。
不是睁大眼,也不是夸张惊讶。只是正在进行的表情停半拍,视线仍停在童锦程脸上,随后眼睑轻轻放松,嘴角出现一个极小、压不住的真实弧度。
真实毛孔、细小绒毛、眼球湿润反光和暖冷混合光线清晰保留。
童锦程保持画外位置,在她反应刚出现时,用@音频1音色自然转回直播状态:
{兄弟们,回来了。}
女主播没有说话。
画面停在她已经明显变软、但仍克制的眼神上结束。
【连续性】
全片只有@图片2定义的一个女主播实例和@图片3定义的一个童锦程实例。人物五官、发型、发量、服装结构和体型在所有景别之间保持稳定。
所有近景和特写继续继承@图片1相同的暖色吊灯方向、窗外冷青环境光、曝光和黑位,不产生独立人像棚拍效果。人物皮肤、头发、服装、桌面和背景必须共享同一个真实光照空间。
所有正反打保持同一摄影轴线和正确眼神方向。每次切镜都由上一句台词、停顿或人物反应驱动。对话节奏保留真实聆听时间,不连续抢台词,不出现双方同时说话口型。
场景画面保持纯真人实景电影质感,手机通知文字以及整条视频的重点字幕均作为后期图文层处理,不改变人物、摄影和表演。