由 MiniMax 出品

MiniMax H3 提示詞

MiniMax 的多模态视频模型——从文字、图片、视频片段或音轨出发,生成自带原生立体声的视频。

video
TITLE: THE LAST WITNESS STYLE: Ultra-photorealistic, high-budget Hollywood crime mystery thriller, 16:9 widescreen, 15 seconds, anamorphic cinema lens, dark blue-grey color grade, dramatic practical lighting, heavy rain, atmospheric fog, realistic reflections, subtle film grain, deep shadows, cinematic contrast, natural motion blur. Premium theatrical movie cinematography. No subtitles, no captions, no watermark. CHARACTER LOCK: Detective Daniel, a 38-year-old man with short wet dark hair, light stubble, sharp tired eyes, wearing a black tailored trench coat over a charcoal shirt. Maintain exactly the same face, hairstyle, clothing and proportions in every shot. LOCATION: Abandoned underground parking garage at midnight, wet concrete floor, flickering fluorescent lights, parked black sedan, distant city lights visible through the entrance. A mysterious evidence envelope lies on the car hood. SHOT-BY-SHOT: SHOT 1 (0:00–0:02): EXTREME WIDE ESTABLISHING SHOT. Camera slowly descends into a deserted underground parking garage. Heavy rainwater drips from the ceiling. Detective Daniel walks alone toward a black sedan. Slow cinematic dolly movement, deep atmospheric tension. SHOT 2 (0:02–0:04): LOW-ANGLE TRACKING SHOT. His polished black shoes step through a shallow puddle. Camera tracks backward as he approaches the car. Water ripples naturally with each step. A distant metallic bang echoes. SHOT 3 (0:04–0:06): TIGHT CLOSE-UP. Daniel notices a sealed evidence envelope on the hood. His hand slowly reaches toward it. Camera pushes in toward his eyes, capturing a subtle expression of suspicion. Fluorescent light flickers once. SHOT 4 (0:06–0:09): MACRO INSERT TO OVER-THE-SHOULDER. He opens the envelope and pulls out a photograph. The photograph clearly shows Daniel himself standing beside the same black car, photographed from behind only moments earlier. His fingers freeze. A slow camera push reveals the photograph. SHOT 5 (0:09–0:12): EXTREME CLOSE-UP. Daniel slowly raises his eyes toward the dark reflection in the car window. In the reflection, a mysterious human silhouette stands directly behind him. The silhouette must remain physically grounded and visible only in the glass reflection. Daniel's breathing becomes audible. SHOT 6 (0:12–0:15): FAST WHIP-PAN TO REVEAL. Daniel turns sharply. The space behind him is completely empty. Camera rapidly pushes toward his shocked face. In the final half-second, a second handprint suddenly appears on the fogged car window beside his reflection. CUT TO BLACK. CAMERA: Smooth controlled dolly, low tracking shots, macro detail, shallow depth of field, realistic handheld movement only during the reveal, cinematic rack focus, natural motion blur, consistent screen direction. AUDIO: Heavy rain ambience throughout, echoing footsteps, distant metallic impact, envelope paper rustle, low sub-bass suspense drone, heartbeat-like percussion during the reflection reveal, sharp cinematic sting at the final handprint, abrupt silence on cut to black. REALISM: Physically accurate rain, water reflections, hand movements, paper interaction, believable facial micro-expressions, consistent character identity, realistic glass reflections and lighting continuity. Every shot must connect naturally as one continuous cinematic sequence. NEGATIVE CONSTRAINTS: No face changes, no duplicate characters, no extra fingers, no deformed hands, no floating objects, no cartoon look, no CGI-looking skin, no artificial camera teleportation, no random scene changes, no flickering faces, no inconsistent clothing, no unreadable generated text, no subtitles, no logos, no watermark.
# MiniMax H3 中文提示词|25秒,《狂飙》高启强&安欣经典对峙对话,中文对白,16:9,24fps,2K output_spec: 画幅16:9,时长25秒,24fps,2K,写实电视剧电影质感,低对比暗调,轻微胶片颗粒,无字幕,无水印,无画面文字,人物面部稳定,动作连贯,压抑拉扯氛围。 integrated_multimodal_description: 室内茶室,深色实木茶桌,一盏暖黄台灯,背景昏暗留白,京海氛围感。高启强身着深色西装,沉稳内敛,眼神复杂;安欣穿黑色便服,脊背挺直,目光坚定。两人隔着茶桌对峙,克制的心理博弈,暗流涌动。 人物设定: 高启强:中年男人,沉稳,眼底藏沧桑,语气从容带试探; 安欣:中年刑警,头发花白,神情执拗,冷静克制,坚守底线。 [Shot 1] 00:00-00:05 景别:双人中景 运镜:固定机位,极慢向前推镜 画面:两人隔着茶桌相对而坐,台灯只照亮上半身,一半脸在阴影里,高启强指尖轻叩茶盏 台词(高启强,平缓):安欣,要是当初我还是卖鱼的,你会把我当朋友吗? [Shot 2] 00:05-00:11 景别:安欣近景 运镜:硬切,定镜 画面:安欣盯着高启强,眉头微收,神色沉重 台词(安欣,冷静):哪有那么多如果,路都是自己选的。 [Shot 3] 00:11-00:17 景别:高启强近景 运镜:硬切,缓慢push in 画面:高启强低头轻笑,抬眼看向安欣,眼神怅然 台词(高启强,低声):我做好人时,坏人欺负我;做坏人,好人又审判我。 [Shot 4] 00:17-00:21 景别:双人近景 运镜:硬切,轻微横向摇镜 画面:安欣身体微微前倾,目光锐利,直面高启强 台词(安欣,语气坚定):黑白有界,法理不会模糊。 [Shot 5] 00:21-00:25 景别:双人特写同框 运镜:镜头zoom in,定格在两人对视的面部 画面:沉默对视,台灯光影在脸上晃动,张力拉满 台词(高启强,轻声):你还是没变。 overall_soundscape: 安静茶室环境底噪,茶盏轻微磕碰声,布料摩擦声,呼吸声,音效克制不抢人声 non_diegetic_music: 低沉压抑的悬疑弦乐,音量压低,留白多,对白间隙缓慢烘托情绪
# 30秒写实电影质感|神龙政变·长生殿逼宫剧情短片 ## 一、整体基础设定 生成一支30秒、中式古装写实电影质感、原生同步中文对白的历史剧情短片。 画幅为宽银幕20:9,写实电影摄影,24fps观感,轻微细腻35mm胶片颗粒,压低数码锐化。 场景设定武周末年洛阳皇宫长生殿内,深色描金梁柱、厚重暗紫色丝绒帷幔、青铜烛台、古朴地砖。拂晓时分,清冷微弱天光从高处窗棂斜照进大殿,烛火摇曳,空气中浮着淡淡的烟尘。刚刚经历杀伐,殿内气氛紧绷肃杀。 整体画面色调克制沉郁:暗紫、古铜、炭灰、褪色朱红,走正史写实风格。杜绝仙侠特效、浮夸影楼古装、现代物件、商业广告质感。 摒弃舞台剧式夸张表演,冲突依靠神态、呼吸、眼神张力、人物气场推进,是老臣以大势规劝女皇还政,不是粗暴武力咆哮。 ## 二、故事内核&表演基调 核心冲突:神龙政变成功诛杀二张之后,八十岁宰相张柬之带领禁军来到长生殿,当面逼迫已经大权旁落的武则天交出皇权、传位太子。一边是耄耋老臣,隐忍多年、意志坚定;一边是叱咤半生、骤然被夺权,从震惊、强作威严,慢慢认清大势已去。 表演基调:张柬之举止恭敬,语气沉稳坚定,有理有据,不带狂妄;武则天从骤然惊醒的错愕、女皇式的威压质问,一步步转为心力交瘁、万般无奈。情绪轨迹:张柬之:沉稳肃穆→恳切强硬;武则天:惊惶震怒→心存侥幸→无力认命。戏剧重心落在权力交接那一瞬间无声的拉扯。 ## 三、人物一致性 ### 张柬之 80岁东亚老年男性,身形清瘦腰背依旧挺直,面容布满深刻皱纹,花白胡须,眼神沉静锐利,自带饱经世事的沧桑。身穿武周时期宰相绛色朝服,官帽规整,衣料厚重垂顺。 声线:苍老沉稳厚重男声,语速不快,字句清晰有力,态度恭谨却立场绝不动摇。 ### 武则天 78岁东亚老年女性,鬓发花白,脸上布满岁月纹路,身着深紫色寝居便式宫装,刚从睡梦中惊起,发髻微乱,褪去平日全副帝王盛装,依旧残存女皇气场,眼神锐利。保留真实皮肤纹理,拒绝过度美颜。 声线:略带沙哑苍老女声,一开始带着帝王的威严怒意,后面声调缓缓下沉,透出疲惫无力。 全局人物约束 两人全程面孔、年龄、发型、服饰保持不变,禁止面部漂移、五官变形,不模仿影视剧中已知演员长相。 严格遵守180度轴线:张柬之位于画面左侧,武则天位于画面右侧,二人相隔约两米,视线对接,站位固定不互换。 ## 四、分镜头时序模块 【镜头一|0.0‑6.3秒】 85mm人像镜头,张柬之胸部以上近景,三分之二侧脸朝向画面右侧武则天。背景虚化晃动的帷幔与青铜烛台,极浅景深。 摄影机近乎完全固定,只保留极其细微的手持呼吸感,无推拉、环绕、升降运镜。 0.0‑2.9秒:老人身姿挺拔,眉头微敛,花白长须随着微弱呼吸轻微起伏,目光稳稳看向女皇,神色肃穆,没有半分张狂。 表情辅助:AU4眉毛轻压,AU23嘴唇收紧。 2.9‑6.3秒:微微躬身,做出臣子觐见姿态,语气平稳却不容退让。 张柬之声色沉稳: “二张谋反作乱,臣等奉太子之令,已将二人诛杀。” 台词说完立刻硬切,无淡入淡出。 【镜头二|6.3‑17.6秒】 硬切至武则天越肩近景,张柬之朝服肩头作为模糊前景占据画面左下角1/5。武则天处在画面中央偏右,胸部以上构图,背景是被风吹动的丝质帷幔。 摄影机基本固定,仅有几乎察觉不到的轻微呼吸晃动;随着女皇心神动摇,镜头极其缓慢微小向前推进。 6.3‑9.6秒:武则天猛地惊起,眉心骤然紧锁,双眼睁大,胸口急促起伏,睡意全无,带着上位者被冒犯的震怒。 表情辅助:AU5上眼睑抬高,AU4眉心下压。 9.6‑13.2秒:强压下慌乱,努力维持女皇威仪,目光锐利盯住张柬之,声音带着惊魂未定的质问。 武则天语气威严,带着怒气: “是谁在宫外作乱?!” 13.2‑17.6秒:听完叛乱已经平定,眼底闪过一丝震动,肩膀微微垮下,内心开始意识到局势已经失控,眼底泛起水光,但还没有落泪。卡点硬切。 【镜头三|17.6‑24.1秒】 硬切回张柬之近景,焦段、光线、机位与镜头一完全一致。武则天的画外音从画面右侧传来。 17.6‑19.8秒:张柬之眼神没有丝毫躲闪,神情依旧平静,没有被女皇的气势震慑。 20.0‑22.0秒:下颌微微收紧,语气恳切,但立场丝毫不松,摆出无可逆转的现实。 22.0‑24.1秒:微微低头行礼,说出最关键的诉求。 张柬之继续沉稳进言: “太子不可再返回东宫,请陛下传位于太子。” 【镜头四|24.1‑30.0秒】 硬切至武则天面部极近特写,相比前一个镜头机位更加靠近脸部。 24.1‑26.6秒:眉毛内侧向上挤压,眼眶迅速蓄满泪水,一滴泪珠缓缓从苍老脸颊滑落,鼻翼微微颤动,嘴唇无力的轻抖。 26.6‑30.0秒:女皇的眼神彻底失去往日锋芒,头颅微微低下,一股深深的无力席卷全身,声音疲惫沙哑。 武则天语气低沉、心力交瘁: “事已至此……便依你们所言。” 台词结束定格武则天落寞失神的面部0.3秒,短片直接收尾,不做淡出。 ## 五、声音设计 原生生成中文对白,区分两位说话人,口型与每一个音节准确同步。 张柬之人声:苍老厚重男声,情绪轨迹:肃穆冷静→有理有据→坚定规劝。语速平缓,不靠音量压迫,依靠话语本身的分量。 武则天人声:苍老沙哑女声,情绪轨迹:惊醒震怒→强撑威严→大势已去、满心疲惫。 保留真实人声细节:沉重呼吸、气息起伏、苍老的颤音,禁止过度降噪。 环境音:长生殿空旷的回声,远处隐约传来甲胄轻微碰撞声,烛火噼啪微弱声响,大殿安静底噪。 禁止背景音乐、配乐、旁白、夸张戏剧音效。 ## 六、影像限制 禁止字幕、水印、logo以及任何可读文字。 禁止额外第三人、场景中途切换、人物服饰发型发生变化、人脸漂移。 禁止美颜磨皮、仙侠滤镜、夸张大幅度肢体动作、嘶吼尖叫、慢动作、环绕镜头、急速推拉。 禁止目光涣散漂移、口型错位、嘴唇粘连、过度虚假流泪。 整体风格写实历史古装,武周服饰器物符合时代特征,不要唐宋服饰混搭,拒绝CG游戏质感。 ## 七、核心主旨重申 短片重点刻画权力交接最后一刻沉重无声的博弈,依靠微表情、眼神、人物气场烘托紧张氛围,不用花哨运镜,还原八十岁老臣张柬之孤注一掷,逼迫一代女皇武则天还政李唐这一历史性瞬间。
【PROJECT】 MiniMax H3, 15 seconds, 16:9. MIDNIGHT GLASS: photoreal live-action electronic-pop performance. @Audio1 is the timing master. Intimate, mysterious, emotionally exposed. Match actual phrases and pulse, not an invented theatrical climax. Eight shots share one coherent location. 【REFERENCES】 @Image1 supplies the woman's recognizable face, long wavy platinum-blonde hair and proportions. Replace the red dress and original stage. @Audio1 supplies the approved 15-second excerpt unchanged in pitch, speed, vocal and arrangement. Use this actual recording, not a replacement singer or newly composed song. One physical woman; reflections are synchronized optical reflections, never additional performers. 【CREATIVE】 Opening emotional hook: she looks from dark glass directly toward the viewer within the first second, as though finally understood. Vulnerability gradually becomes quiet resolve. Connected route: smoked-glass corridor → luminous doorway → reflective chamber perimeter → central performance light. Reveal adjacent architecture rather than teleporting between unrelated sets. Her face, breath and singing lead the film; effects remain subordinate. 【MUSIC / OUTFIT / PERFORMANCE】 Preserve actual vowels, consonants, breath gaps, sustained notes and instrumental textures. No invented bass drop, silence, shouted high note, lyrics or orchestral stop. Mouth articulation matches the uploaded vocal. No dialogue or additional singers. One midnight-plum satin gown: asymmetric neckline, fitted bodice, defined waist, flowing ankle-length skirt, modest side slit. Small silver drop earrings, discreet stable heels, no necklace. Same gown in every angle. Satin catches soft silver highlights in blue light; hair remains platinum and wavy. No red reference dress or costume changes. Use gaze turn, contact/release, doorway crossing, diagonal walk, pivot, open-forearm gesture and two still points. Every step has grounded weight transfer. Turns originate through shoulders and hips; skirt and hair follow with natural inertia. Controlled emotional singing, not dance choreography. No compulsory tears or exaggerated jaw opening. Preserve natural even skin tone. Face and mouth stay clear. 【VISUAL / TYPE】 Midnight blue, deep plum, black glass, restrained violet haze and silver highlights. Soft side light shapes realistic skin; doorway edges provide depth. Dry polished floor, not water. Readable dark areas, controlled highlights, clean cinematic photography without grain. Elegant condensed uppercase sans, moderate weight, generous tracking. Exact editorial words only: “WITHIN”, “OPEN”, “DEEPER”; not lyric quotations. Each appears whole and readable before moving. Layer A: architectural hero word. Layer B: faint glass reflection. Layer C: small foreground echo. Never obscure eyes or mouth. Methods: glass-edge occlusion, doorway wipe, focus transfer, matched-turn cut, brief text-only reflection offset. Each resolves promptly to clarity. No independent human reflections, face doubles, paper collage, halftone or random glitch. Clean motivated cuts, not soft dissolves. Shot boundaries below may shift slightly to nearby real phrase accents. 【SHOT 1 — 0.0–1.8s】 Close shot, slow lateral track in corridor. Gaze-turn action: begin with her face visible in three-quarter profile, already singing. Eyes turn first, then head, meeting the viewer within one second. Shoulders remain quiet; hair settles after the turn. Reflection follows naturally. The hook is a vulnerable, searching look, not a beauty pose or empty establishing view. Typography “WITHIN”: Layer A silver word behind shoulder; Layer B faint reflection below; Layer C thin echo at extreme left. Keep mouth unobstructed. Rhythm / Cut: reveal on actual opening phrase; passing glass edge motivates the next angle. No fabricated sound accent. 【SHOT 2 — 1.8–3.6s】 Medium side view tracking parallel to corridor wall. Contact/release action: fingertips touch stationary glass, then separate clearly as she shifts weight away. Gaze follows her reflection briefly before returning forward. Feet remain planted during contact; camera shows hand-to-glass separation. Singing continues naturally. Typography “OPEN”: Layer A beyond the glass, fully readable; Layer B dim reflected duplicate; Layer C narrow lower-edge echo. Nothing overlays her lips. Rhythm / Cut: focus transfers from hand to face; cut near an actual syllabic accent. No manufactured drum hit or magic movement through glass. 【SHOT 3 — 3.6–5.5s】 Medium-full view, camera retreats through luminous doorway. Threshold action: two purposeful steps enter the chamber; forearm briefly brushes the solid frame and releases. Body clears the opening normally. Weight transfers through grounded shoes; skirt follows behind and settles. Camera maintains comfortable separation. Typography “DEEPER”: Layer A distant chamber wall; Layer B subdued floor reflection; Layer C small foreground echo briefly hidden by skirt. Rhythm / Cut: doorway edge motivates occlusion cut, with continuous recorded singing. Show architectural continuity and no teleportation. Hair and gown never cross solid surfaces. 【SHOT 4 — 5.5–7.3s】 Wide-to-medium lateral track around chamber perimeter. Diagonal traversal: she walks across a silver light band, turning torso toward camera while feet follow the clear floor route. Show full silhouette and broader satin movement. One hand opens at her side then relaxes. Resolve with a grounded stop. Typography “OPEN”: Layer A far wall; Layer B soft glass reflection; Layer C small vertical edge repeat. Whole word reads before its perspective changes. Rhythm / Cut: movement follows existing pulse wh
15-Second 1930s American Melodrama, Close-Up Focus, English Dialogue Title: Rainy Night at the Logan House Genre: 1930s American romantic melodrama, nostalgic old Hollywood vintage TV film texture, quiet emotional tug-of-war Duration: 15 seconds Aspect Ratio: 16:9 | 24fps feel Visual Style: vintage 35mm film grain, warm brown faded tone, soft gray-blue shadows, low-contrast old Hollywood table lamp lighting Setting: Wealthy American living room, 1936, night. Fringed table lamp as sole key light, rain streaking window, mahogany furniture, velvet curtains. No modern items. Core Conflict: Vulnerable woman uses tearful soft words to emotionally pressure the man, he sinks into guilt and self-doubt. No yelling, all tension carried by close-ups and reverse cuts. Character Lock:Female lead: late 20s, elegant 1930s bias-cut dress, soft waved hair, teary watery eyes, pale face. Voice soft, breathy, with restrained sorrow and quiet emotional pressure. Male lead: early 30s, tailored dark 1930s suit, shirt & tie. Voice deep, hesitant, guilty, lost for words. Blocking: She sits left on leather sofa; he stands front-right of sofa. Fixed 180° axis, eye contact locked. Shot Breakdown 00:00–00:03|Opening: Female face close-up Table lamp side light half-shadows her face. Eyes brimming with tears, lips trembling, staring steadily at him. Rain patters on glass. 00:03–00:07|Reverse cut, male close-up + her off-screen line He listens, brow furrowed, uneasy. Her soft tearful voice: “If you cared for me, you wouldn’t hurt me like this.”Cut back to her close-up, glistening tearful eyes. Hard cut, no dissolve. 00:07–00:11|Male close-up Adam’s apple bobs, he struggles to reply, voice low and shaky: “That’s not what I meant… I never wanted to hurt you.” 00:11–00:15|Final two-shot close-up, static hold ending Cut back to her face, slight sad head shake. He drops his gaze, overwhelmed by guilt. Rain hums in background. Frame freezes, end directly, no fade, no black screen. Sound Rules English dialogue, strict lip sync matching lines. Voice tone: female soft breathy sob; male hesitant, weary, guilty. Only ambient sound: rain tapping window, faint fabric rustle, quiet breathing, room natural reverb. NO music, BGM, subtitles, narration, extra sound effects. Visual Constraints No text, watermarks, logos. No over-smoothing / over-beautification. No wild gestures, screaming, fancy camera movement, slow motion, flashy transitions. Keep facial features consistent, no face distortion. Core Theme: Gentle emotional coercion. She hides pressure behind fragility and tears, leaving him trapped in quiet guilt.
基础设定:1940 年代民国上海,低饱和复古胶片,王家卫式克制忧郁,全程**特写、近景为主**,无大全景,无字幕、无配乐,仅环境音效,24fps,稳镜,微表演,口型同步。 人物: 女主:墨绿色暗纹旗袍,复古盘发,眉眼隐忍,含泪但不哭嚎 男主:深色复古西装,油头,神色沉痛压抑,克制不崩溃 剧情:乱世之下被迫永别,安静告别,没有激烈哭喊,痛都藏在眼底 【00:00‑00:03|开场女主面部特写(硬性开场)】 女主脸部特写,柔光一半落在脸颊,另一半沉在阴影里,眼底蓄泪,睫毛微颤,嘴唇轻轻抿紧,压抑着情绪,几乎不动。 【00:03‑00:09|正反打切男主近景,对白】 平稳切男主半身特写,他望着女主,声音低沉沙哑,克制颤抖:“这一别,恐怕再也见不到了。” 微表情:眉心轻蹙,眼底发红,不落泪,肩膀没有剧烈抖动。 【00:09‑00:15|切回女主特写,对白】 切回女主面部特写,泪水在眼眶打转,没有滑落,声音轻得像叹息:“我知道。你要好好活下去。” 【00:15‑00:21|交替正反打特写,无声对视】 快速柔和正反打,轮流给到男主、女主的眼部特写,两人静静凝望,千言万语都说不出口,只有眼底翻涌的不舍与绝望,无肢体接触。 【00:21‑00:25|女主侧脸特写收尾,定格】 固定女主侧脸特写,她缓缓闭上眼,画面安静定格,不淡出、不黑屏。 【声音设计】 对白口型精准对齐。环境音效:远处隐约的风声、老上海街道微弱的车马底噪,一声极轻的布料摩擦声。禁止 BGM、旁白、字幕。 【影像约束】 35mm 复古胶片颗粒,低饱和,柔和漫射柔光,民国暗调美学;拒绝高饱和、美颜、夸张表情、快速摇晃镜头、大幅度动作;全程以面部、眼部特写优先,镜头稳定,只做极轻微呼吸感。 【主旨】乱世民国,无声的生离死别,痛而克制,一别永年。
Create a 15-second funny, high-quality 3D animated cartoon video in a colorful, Pixar-inspired family animation style. Characters: 1. A cute, chubby little boy, around 3 years old, with big expressive eyes, funny facial expressions, and adorable baby voice. 2. A cute fluffy orange-and-white cat. 3. A loving mother sitting on the living room sofa. 4. A funny father sitting beside the mother, watching everything. Scene 1 (0–4 seconds): The little boy is playing with the fluffy cat on the living room floor. He tries to take the cat's toy, but the cat playfully grabs it first. The boy looks shocked and says in a cute English voice: Boy: "Hey! That's MY toy!" Scene 2 (4–7 seconds): The cat gently taps the boy's hand with its paw and runs away. The boy gets dramatically upset, starts crying loudly, and says: Boy: "Mooooom! The cat is fighting with me!" Scene 3 (7–11 seconds): The little boy runs toward his mother with funny crying expressions, pointing at the cat. The mother looks surprised and asks: Mother: "Oh honey! What happened?" Boy: "She started it! I did nothing!" Scene 4 (11–15 seconds): The father starts laughing loudly. The cat calmly sits on the sofa looking innocent. The boy suddenly stops crying, points at the cat, and says: Boy: "Daddy! Don't laugh! She's a bad kitty!" The mother laughs, the father laughs even harder, and the cat gives a funny innocent look toward the camera. Visual Style: Beautiful 3D cartoon animation, adorable characters, vibrant colors, realistic cartoon movements, expressive faces, cinematic lighting, detailed cozy family living room. Audio: All dialogues must be in clear English with cute child voices, funny emotional expressions, natural lip-sync, mother and father reactions, cat meowing, and funny cartoon sound effects. Important: Keep the same character designs throughout all scenes. Fast-paced comedy, smooth transitions, perfectly synchronized English dialogue, vertical 9:16 format, exactly 15 seconds long.
THE WRONG TABLE — EPISODE 1 — CLIFFHANGER FREEZE-FRAME ENDING CRITICAL: NO ON-SCREEN TEXT OR SUBTITLES. No captions, names, titles, logos, or watermarks. REFERENCE IMAGES: — CHEN YICHEN: Use the provided restaurant image of the male lead as the ONLY visual reference for the male character. Strictly preserve his original facial appearance, hairstyle, navy-blue suit jacket, light-blue shirt, calm expression with a slight smile, and the original restaurant background, lighting, and camera angle. Do not redesign the character. — SHEN RUOXI: Use the provided shocked close-up image of the female lead as the ONLY visual reference for the female character. Strictly preserve her original facial appearance, long hair, gold hoop earrings, dark-blue silk blouse, widened eyes, slightly parted lips, and the original background, lighting, and camera angle. Do not redesign the character. integrated_multimodal_description: 9:16 vertical format, modern Chinese urban romantic short drama, cinematic photorealistic live-action visuals, classic Chinese short-drama suspenseful freeze-frame ending. The entire sequence uses ONLY the two original character images from and . Do not create new camera shots or change the characters' faces, expressions, clothing, compositions, or backgrounds. Create suspense through frozen images, rapid cuts, white flashes, soft misty light blooms, and fade-out effects. The character appearance order must strictly be: Shen Ruoxi → Chen Yichen → Shen Ruoxi. [Shot 1] At 00:00.000, immediately show the shocked facial close-up of Shen Ruoxi from . Her eyes are widened and her lips are slightly parted, preserving the exact shocked expression from the original image. The image freezes completely. No blinking, no head movement, and no lip movement. A soft, misty white glow gradually appears around the edges of the frame, followed by a rapid white flash at 00:00.850. [Shot 2] At 00:01.000, immediately after the white flash, hard cut to the original restaurant image of Chen Yichen from . Preserve his original seated posture, subtle smile, clothing, and background. The image remains completely frozen, creating a strong contrast with the female lead's shocked expression. Hold briefly, then introduce another rapid white flash. [Shot 3] At 00:02.000, immediately after the flash, cut back to the original shocked close-up of Shen Ruoxi from . The image instantly freezes, preserving her exact original facial expression and composition. Add a subtle, soft white misty glow around the edges, a hazy light bloom, and a brief ghosted afterimage overlay to create the suspenseful atmosphere of a classic Chinese short-drama cliffhanger ending. Her face must remain clearly visible at all times and must not be obscured by the white glow. [Shot 4] At 00:03.000, continue holding the completely frozen close-up of Shen Ruoxi. The soft white glow briefly intensifies, then gradually diminishes. During the final half-second, smoothly fade the image to pure black. End at 00:04.000. VISUAL CONTINUITY: Strictly preserve the characters' identities, original facial structures, hairstyles, clothing, natural skin textures, lighting, backgrounds, and camera compositions from the two reference images. All character images must remain completely motionless freeze-frames. Only editing transitions and subtle optical effects are permitted. Do not generate new actions, change facial expressions, animate lips, introduce additional characters, change locations, rotate the camera, zoom in, or zoom out. No subtitles, no text, no graphic titles, and no watermarks. overall_soundscape: Begin with a short, deep cinematic suspense impact sound. Each rapid white flash is accompanied by a subtle, clean air-whoosh sound. During the final freeze-frame, maintain a brief lingering low-frequency suspense resonance that quickly disappears as the image fades out, ending in complete silence. No character dialogue, no voiceover, and no restaurant conversations. non_diegetic_music: None. Use only brief suspense impact sound effects and white-flash transition sounds. No continuous background music.
TITLE: GTA VI — LUCIA'S NIGHT RUN STYLE: Ultra-realistic AAA open-world video game gameplay inspired by Grand Theft Auto VI. 16:9 landscape, exactly 15 seconds, photorealistic Rockstar-style game-world atmosphere, modern Vice City at night, Florida-inspired Leonida environment, realistic character animation, detailed environmental textures, dynamic lighting, realistic reflections, premium next-generation console graphics. This must look like actual third-person free-roam gameplay, NOT a cinematic movie trailer. CHARACTER LOCK: Lucia Caminos from GTA VI. Preserve her recognizable official character appearance, facial structure, hairstyle, skin tone and body proportions throughout every shot. She wears a fitted casual top, denim shorts and sneakers. Same outfit in every scene. LOCATION: A busy Vice City neighborhood at night. Small American convenience store with glass entrance doors, brightly lit shelves, refrigerators, checkout counter, colorful neon signs, palm trees, parked cars, wet asphalt and a black sport motorcycle parked directly outside the store. SHOT 1 (0:00–0:03) — THIRD-PERSON GAMEPLAY CAMERA. Camera positioned slightly behind Lucia as she enters the convenience store. Automatic glass doors slide open. She walks naturally toward the refrigerated drinks section. Bright fluorescent store lighting contrasts with the neon-lit street outside. A few customers casually move in the background. SHOT 2 (0:03–0:06) — OVER-THE-SHOULDER GAMEPLAY. Lucia opens the refrigerator, takes a cold bottled drink and grabs a packet of chips from a nearby shelf. Close-up of her hand placing both items on the checkout counter. A cashier scans the products. Natural hand movements, believable object interaction, realistic refrigerator lighting. SHOT 3 (0:06–0:08) — MEDIUM FOLLOW SHOT. Lucia completes payment, picks up her small shopping bag and turns toward the entrance. The automatic doors open. Camera smoothly follows her outside. The warm neon signs and colorful Vice City street become visible. SHOT 4 (0:08–0:11) — THIRD-PERSON EXTERIOR GAMEPLAY. Lucia approaches her black sport motorcycle parked beside the sidewalk. She places the shopping bag securely in the motorcycle's storage compartment, takes the handlebars and swings one leg naturally over the seat. Realistic body weight and motorcycle suspension. SHOT 5 (0:11–0:13) — CLOSE SIDE ANGLE. Lucia starts the engine. Headlight switches on, engine vibrates realistically, rear tire moves slightly. She accelerates smoothly away from the sidewalk and enters the street. SHOT 6 (0:13–0:15) — THIRD-PERSON CHASE CAMERA. Camera follows behind the motorcycle as Lucia rides through a neon-lit Vice City street. Palm trees, colorful storefronts, moving traffic, wet road reflections and distant skyscrapers create a believable open-world gameplay atmosphere. End while she continues riding into the city. GAMEPLAY CAMERA: Authentic GTA-style third-person follow camera, smooth player-controlled movement, slight camera lag during acceleration, realistic steering and turning, consistent camera distance, natural vehicle physics. No random cinematic camera cuts. AUDIO: Automatic store door chime, refrigerator hum, footsteps, product packaging rustle, checkout scanner beep, muffled store ambience, motorcycle ignition, realistic engine revving, tire movement and distant city traffic. Subtle Vice City-inspired original background radio music. VISUAL QUALITY: High-detail next-generation open-world graphics, realistic character animation, detailed store interiors, natural facial expressions, accurate hand-object interaction, physically correct motorcycle movement, dynamic shadows, realistic wet asphalt reflections, consistent lighting and environmental continuity. NEGATIVE CONSTRAINTS: No character face changes, no outfit changes, no duplicate Lucia, no extra fingers, no deformed hands, no floating shopping bag, no teleportation, no motorcycle morphing, no disappearing objects, no unnatural walking, no sliding feet, no unrealistic acceleration, no random camera cuts, no cartoon graphics, no fake subtitles, no watermark, no unrelated logos.
# MiniMax H3 中文提示词|25秒,《大话西游》经典“爱你一万年”片段,中文对白,16:9,24fps,2K output_spec: 画幅16:9,时长25秒,24fps,2K,电影画面质感,无字幕,无水印,无画面文字,人物面部稳定,五官清晰自然,动作连贯,镜头节奏克制。 integrated_multimodal_description: 《大话西游之大圣娶亲》经典喜剧电影质感,荒凉山寨石洞内部,石壁粗粝,烛火摇晃,低饱和复古色彩,轻微胶片颗粒,苍凉又深情的氛围。紫霞仙子执剑与至尊宝近距离对峙,画面中只有紫霞仙子、至尊宝、长剑、火把与石壁背景。 人物设定: 紫霞仙子:年轻女子,古装女侠造型,眼神倔强又带伤感,手持长剑,剑身横在至尊宝面前,情绪从戒备、嘲讽逐渐转为震撼与含泪。 至尊宝:年轻男子,略带不羁的古装造型,头发微乱,表情从无奈转为认真、悲怆,最后深情告白。 [Shot 1] 00:00-00:05 景别:中近景双人同框。 运镜:固定机位,轻微手持呼吸感。 画面:紫霞仙子持剑横在至尊宝颈前,两人站在石洞中央,烛火从侧后方照亮人物轮廓。紫霞盯着至尊宝,剑尖停在他胸前,动作克制。 台词(紫霞,冷声带笑):“你以为我不敢杀你?” [Shot 2] 00:05-00:11 景别:至尊宝近景。 运镜:硬切,镜头缓慢push in。 画面:至尊宝看着紫霞,表情由吊儿郎当变得认真,眼眶微红,背景烛火虚化成暖光斑。 台词(至尊宝,语气郑重):“曾经有一份真诚的爱情放在我面前,我没有珍惜。” [Shot 3] 00:11-00:17 景别:双人中近景。 运镜:硬切,轻微缓慢推镜。 画面:紫霞握剑的手微微发紧,至尊宝站在她面前,没有后退,两人目光牢牢相接,石洞内火光轻晃。 台词(至尊宝,声音低沉):“等我失去的时候,我才后悔莫及。” [Shot 4] 00:17-00:21 景别:紫霞特写。 运镜:硬切,定镜。 画面:紫霞眼神动摇,剑尖缓缓下垂,嘴唇轻颤,眼底泛起泪光,原本的倔强被击中。 台词(至尊宝,画外音):“人世间最痛苦的事,莫过于此。” [Shot 5] 00:21-00:25 景别:双人近景。 运镜:镜头极慢zoom in,收束在两人对视。 画面:至尊宝看着紫霞,紫霞的剑已经放下,两人之间空气安静,烛火映在脸上,画面深情定格。 台词(至尊宝,轻声):“如果非要加一个期限,我希望是——一万年。” overall_soundscape: 石洞空旷风声,火把燃烧声,长剑轻微金属声,衣料摩擦声,人物呼吸声,环境底噪自然,不盖对白。 non_diegetic_music: 低沉忧伤的二胡与箫声,旋律缓慢苍凉,音量低,只在对白空隙增强情绪,不抢人声。
【PROJECT】 MiniMax H3, 15 seconds, 16:9. MIDNIGHT GLASS PART 2: photoreal live-action rock-stage performance. Epic concert lighting, emotionally intense singing and bold layered typography. Audio is the master clock. Eight shots in one stage; no unrelated scene changes. 【REFERENCES】 @Image1 supplies the rock stage, leather outfit, red lipstick, smoky eyeliner, microphone and two guitarists. @Image2 locks the lead woman's recognizable face, facial proportions, hairline and platinum-blonde waves only, not its red dress. @Audio1 is the approved first 15 seconds of audio2. Preserve recorded voice, instruments, pitch, tempo and phrasing; no replacement singer or music. 【CREATIVE】 Opening hook: immediate emotional eye contact and enormous stage lettering. Vulnerability grows into grounded rock power with the actual recording. Connected zones: microphone position, front apron, side-monitor lane and central spotlight. Exactly one lead singer plus two guitarists behind her, left and right. Preserve stage geography. 【MUSIC / FACE / OUTFIT】 Face identity has priority: preserve @Image2's recognizable face at every angle, natural skin texture and even complexion. Long platinum waves, red lipstick and smoky eyeliner remain consistent. No beautification that changes identity. Same fitted black leather biker jacket, secure black stage top, high-waisted leather trousers, silver hardware and earrings throughout. One wired handheld mic stays in her right hand. Left hand gestures freely. Cable follows gravity and a safe floor route. Mouth shapes match actual recorded syllables and breath gaps. Sustained notes have natural jaw movement and held posture. No invented screaming or lyrics. Retain recorded backing vocals without adding visible singers. Gaze lift, stance shift, diagonal walk, shoulder turn, arm release and two still points provide varied performance. Grounded steps, coherent joints, jacket and hair inertia. Fan airflow keeps hair clear of mouth. Guitarists play with plausible hands and attached straps. 【VISUAL / TYPOGRAPHY】 Rear LED wall, truss, amplifiers, speaker cabinets, pedalboards, wedge monitors, cables and unoccupied drum kit. Blue-violet beams, crimson rear accents, soft white face key and restrained haze. Realistic leather, skin and metal, controlled exposure, no grain. Typography is a PRIMARY element, not subtle decoration. Ultra-black condensed uppercase grotesk, tight tracking, ivory hero letters with silver/crimson echoes. Exact phrases: “I'VE BECOME SO NUMB” “WITHOUT A SOUL” “FIND IT THERE” “LEAD BACK HOME” Last two are editorial phrases, not compulsory sung lyrics. Never alter recorded vocal to match text. Layer A: huge complete rear-screen phrase in readable lines. Layer B: smaller silver side-screen echo. Layer C: lower-foreground word fragment. Keep mouth and eyes clear. Full words hold before transforming. Use text-only offsets, light sweeps, split panels and clean cuts. Restore clear text and one sharp singer. No karaoke strips or gibberish. Times can shift slightly to actual vocal accents. 【SHOT 1 — 0.0–1.8s】 Close-up at mic, slow push. She sings and raises gaze into direct eye contact within first second, left hand near sternum. Face immediately recognizable. Typography “I'VE BECOME SO NUMB”: Layer A full huge phrase beside silhouette; Layer B silver side echo; Layer C NUMB low edge. Rhythm / Cut: opening vocal reveals complete wording, then gaze motivates clean cut. No haze obscuring text. 【SHOT 2 — 1.8–3.6s】 Medium apron shot, lateral track. Strong planted stance, shoulders open, left arm extends once, right mic steady. Hair follows airflow; guitarists stay behind. Typography same phrase: Layer A full rear words; Layer B text-only offset; Layer C NUMB below waist. Rhythm / Cut: real phrase emphasis aligns letters. Restore offsets promptly, lips remain clear. 【SHOT 3 — 3.6–5.5s】 Medium-full diagonal track beside monitors. Two purposeful steps toward central light, safe feet and natural cable slack. Jacket follows torso; no equipment intersection. Typography “WITHOUT A SOUL”: Layer A rear phrase; Layer B narrow side echo; Layer C SOUL foreground. Rhythm / Cut: actual rise activates crimson accent. Monitor edge motivates cut, no invented drop or hidden mouth. 【SHOT 4 — 5.5–7.3s】 Wide stage shot, restrained arc. Singer reaches center, opens left arm, mic stays right. Both guitarists and stage equipment visible. End planted. Typography “WITHOUT A SOUL”: Layer A immense full phrase; Layer B side panels; Layer C low SOUL. Rhythm / Cut: actual musical accent widens beams and aligns type. Stable stage, no strobe barrage. 【SHOT 5 — 7.3–9.1s】 Intimate close-up, nearly locked. First still point: concentrated singing, natural breath, searching gaze. Face exactly follows identity reference; mic does not hide lips. Typography “FIND IT THERE”: Layer A full phrase beside face; Layer B faint silver repeat; Layer C THERE bottom edge. Rhythm / Cut: light sweep reveals then holds words. Editorial text does not change lip-sync. No forced tears. 【SHOT 6 — 9.1–11.0s】 Medium-full near monitor, low undistorted view. Shoulder-led quarter turn then grounded return front. Left hand briefly contacts existing stand and releases, right hand keeps mic. Hair, jacket and cable settle. Typography “FIND IT THERE”: Layer A rear phrase; Layer B registration echo; Layer C THERE exits sideways. Rhythm / Cut: actual phrase resolves turn. Brief guitar-detail split panel returns to one stage view, no body duplication. 【SHOT 7 — 11.0–13.0s】 Medium track toward front light. One deliberate step, chin lift and left-arm release wi
Duration: 15 seconds Aspect Ratio: 16:9 Frame Rate: 24fps Visual Style: 1940s Hollywood black-and-white film, high-contrast chiaroscuro, vintage Warner Bros. studio texture, soft cigarette smoke, subtle film grain, restrained romantic melancholy. Setting: Rick’s Café Americain in Casablanca, at night. Dim smoky interior, dark wood bar, faint neon glow through the window, a half-empty glass on the counter. Core Tone: Bittersweet farewell, quiet tension between memory and duty. Close-ups and shot/reverse shot. Open with Ilsa’s face close-up. Character Lock:Ilsa: Late 20s, elegant 1940s dress, soft wavy hair, teary but restrained eyes. Voice low, tender, regretful. Rick: Late 30s, dark suit, white shirt, stubble, guarded expression. Voice deep, tired, resigned. Blocking: They stand near the bar, a short distance apart. Maintain 180° axis and locked eye contact. Shot Breakdown 00:00–00:03|Opening: Ilsa Face Close-Up Soft shadow splits her face. Her eyes glisten with tears, but she does not cry. She looks directly at Rick, breathing faintly. Cigarette smoke drifts slowly behind her. 00:03–00:07|Reverse Shot, Rick Close-Up + Ilsa Off-Screen Line Rick watches her, his jaw tight. Ilsa speaks softly: “We will always have Paris.” Cut back to her close-up. Hard cut. No dissolve. 00:07–00:11|Rick Face Close-Up His eyes are tired and bitter, but not angry. He says quietly: “Maybe not tonight. But we will have it.” 00:11–00:15|Final Two-Shot Close-Up, Static Hold Cut back to Ilsa. She gives a faint, sad smile. Rick turns his gaze toward the dark doorway. The frame holds still. End directly. No fade out. No black screen. Sound Design English dialogue, strict lip sync. Ilsa: low, tender, regretful female voice. Rick: deep, tired, resigned male voice. Ambient sounds only: faint distant jazz club noise, quiet bar room hum, soft breath, glass clink once. No music, no score, no subtitles, no narration, no added sound effects. Visual Constraints No text, logos, watermarks. No over-smoothing. No modern clothing or props. No yelling. No exaggerated gestures. No circling camera, slow motion, flashy transitions. Keep faces stable and natural. Authentic 1940s black-and-white Warner Bros. film style. Core Theme Memory is the only thing they can keep. A quiet, bittersweet farewell between two people who still love each other but cannot stay.
生成一支25秒、怀旧老电视剧胶片质感、原生同步对白民国爱情剧情短片。画幅16:9,24fps观感,复古35mm胶片颗粒,色调温润偏旧。民国上海夜晚陆家客厅,台灯柔光,窗外细雨,琼瑶式情感拉扯对手戏。全程无字幕、无音乐、无配乐,保留真实现场环境音效,大量面部特写+正反打,开场直接给到女主面部特写。 【全局视听环境设定】 场景民国陆家西式客厅,布艺沙发、琉璃台灯、雕花木家具、薄纱落地窗帘。雨夜,雨水轻敲玻璃窗,室内仅一盏台灯做主光源,光影柔暗。整体色调暖棕、灰蓝、米黄,复古旧剧色彩。杜绝现代器物、鲜亮滤镜、广告打光。 【戏剧内核&表演总原则】 女主以委屈深情的姿态,用情感话术裹挟男主,情感绑架式对话,表面柔弱伤感,言语层层施压,男主陷入自我怀疑,内心痛苦挣扎。多用面部特写、眼神停顿、呼吸留白,正反打交替切换,不歇斯底里嘶吼,是安静、绵密的情感施压。 【人物一致性|全程锁定,禁止变动】 女主(20岁左右,漂亮):精致五官,民国素色旗袍,眼眸含水雾,面色苍白,语气柔弱哀伤,话语暗藏情感裹挟,眼眶盈盈含泪,看着楚楚可怜。人声清亮柔软,带着哽咽感,绵而有压迫力。 男主(20多岁,帅气):民国西装,衬衫领带,面容俊朗,眉头紧锁,神情愧疚纠结,被话语击中后陷入自我怀疑,眼神慌乱痛苦。人声磁性青年男声,语气迟疑、无力。 硬性约束:人物五官、发型、服装全程统一,禁止面部漂移畸变,拒绝过度美颜磨皮,保留皮肤自然纹理。 空间站位:女主坐沙发左侧,男主站沙发右前方,严格180°轴线,视线互相锁定。 【镜头一|0.0—3.0秒|开场女主面部特写】 开篇直接给到女主面部特写。台灯侧光打在脸上,眼中蓄满泪水,嘴唇微微颤抖,目光牢牢锁住画外男主,呼吸轻轻发颤。窗外雨声持续。 【镜头二|3.0—8.5秒|正反打,男主近景】 切男主近景正反打。他站在原地,眉头紧锁,眼神不安,静静聆听。 3.0—5.0秒:男主望着女主,面露不忍,内心开始动摇。 5.0—8.5秒:女主画外轻柔含泪开口: “如果你真的爱我,你就该懂我的痛。你不肯体谅,是不是根本就没有那么在乎我?” 表情辅助(女主,切回特写):泪光晃动,眉眼哀伤,语气柔弱,无声施加情绪压力。硬切,无淡入淡出。 【镜头三|8.5—14.0秒|切男主面部特写】 男主面部特写,听完这句话,喉结滚动,眼神慌乱,下意识想要辩解却又卡住。 8.5—11.0秒:男主嘴唇微张,神色愧疚。 11.0—14.0秒:男主声音迟疑无力: “我不是不在乎,只是……这件事并非你想的那样。” 【镜头四|14.0—19.0秒|切回女主特写,继续情感施压】 女主特写,泪水在眼眶打转,微微摇头,语气愈发凄楚。 14.0—16.0秒:女主轻轻吸气,眼神带着受伤的失望。 16.0—19.0秒:女主柔声说道: “所有伤害我的理由,你都能找到借口。原来在你心里,我永远排在最后。” 【镜头五|19.0—25.0秒|双人近景对峙收尾】 双人近景,正反打收束。女主含泪凝望,男主神情痛苦、自我怀疑,无力反驳。 19.0—22.0秒:男主垂眸,肩膀微微下沉,被情绪裹挟陷入愧疚。 22.0—25.0秒:台灯光影摇曳,窗外雨声沙沙,画面静态定格,直接结束,不黑屏、不淡出。 【声音设计|严格执行】 1.原生中文对白,口型与音节精准同步,男女声区分清晰,禁止口型错位。 2.女主人声:清亮柔软,带着淡淡的哽咽,外表哀伤,话语绵密有情绪压迫感。 3.男主人声:磁性青年男声,迟疑、疲惫,充满愧疚无力感。 4.保留现场环境音效:淅沥雨声、雨打玻璃声、细微衣料摩擦声、人物轻呼吸声、房间微弱空间回响。 5.硬性禁止:背景音乐、配乐、旁白、额外特效音效。只保留对白+现场环境音。 【影像限制|全面避错】 禁止字幕、水印、Logo、任何可读文字;禁止现代物件;禁止过度美颜磨皮、过度锐化;禁止大吼嘶吼、大幅度夸张肢体;禁止环绕镜头、快速推拉、慢动作、花哨转场;禁止五官崩坏、面部漂移;禁止高饱和艳丽滤镜。坚持民国老电视剧复古胶片质感。 【核心主旨】 用密集特写正反打,呈现典型琼瑶式情感绑架。女主以柔弱哀伤的外表,用悲情话术不断制造愧疚感,让男主陷入自我怀疑,温柔外表之下是不动声色的情感裹挟。
Create a 15-second live-action Korean fashion editorial. 16:9 landscape, 24 fps, exactly nine full-frame shots. Keep the model’s face and key actions within a centered 9:16 safe crop. Character Use @Image1 as the only identity reference. Feature one adult Korean woman throughout. Preserve her face, skin tone, long dark-brown waves, wispy bangs, and natural proportions. Do not copy the reference outfit, pose, or background. Visual style A red fashion magazine in motion. Pair realistic skin and directional studio lighting with crisp 2D graphics, cool-white negative space, and a palette of scarlet, burgundy, black, and white. Use one leading graphic per shot: a checkerboard strip, a fine black grid, or a burgundy diagonal panel. Keep graphics away from her eyes. Wardrobe Shots 01–03: Fitted black-and-white top and high-waisted burgundy skirt. Shots 04–06: Scarlet cropped blazer, black high-waisted skirt, and tall boots. Shots 07–09: Asymmetric burgundy dress and silver earrings. Change outfits only on the hard cuts into shots 04 and 07. No wardrobe morphing within shots. Shot sequence 0.0–1.6s | Beauty close-up. She lifts her gaze as a red checkerboard strip slides behind her shoulder. 1.6–3.2s | Full body. She takes half a step and looks back as a burgundy panel slides in. 3.2–4.8s | Seated medium shot. She lowers her hand from her jaw as a fine black grid rises behind her. 4.8–6.4s | Half-body. In the scarlet blazer, she brushes her hair back as a checkerboard strip sweeps behind her. 6.4–8.0s | Low-angle medium shot. She settles one hand on her waist as burgundy diagonals close in from the sides. 8.0–9.6s | Locked close-up. She lowers her chin, then meets the lens. A fine grid stops behind her shoulder. 9.6–11.2s | Full body. In the burgundy dress, she turns and rests one hand on a rail. A checkerboard strip follows the rail’s line. 11.2–12.8s | Shoulder close-up. She adjusts one earring as the graphics pull back into a single red panel. 12.8–15.0s | Final cover. She turns to face the lens. Frame her right of center with clean title space on the left. Hold completely still for the final 0.5 seconds. Music and finish Use a 120 BPM K-pop fashion-pop instrumental with no vocals. Time the graphics to the beat or her gestures. Add CRIMSON INDEX and ISSUE 01 in post for clean, readable type. Keep the movement natural and restrained. No dancing, extra people, identity drift, storyboard grids, or fade to black.
Create a 15-second hyper-realistic video set in Tokyo at night. OUTPUT REQUIREMENTS Aspect ratio: 3:4 vertical Target dimensions: 552 × 736 pixels Duration: 15 seconds Video model: MiniMax H3 Max Do not increase the resolution beyond 552 × 736 simply to improve quality. Do not create multiple video variations. Do not create previews, alternate takes, upscales, or additional assets unless they are absolutely required to complete the requested video. CHARACTER The protagonist is Woman 1, a stylish Japanese woman in her mid-to-late 20s with model-like features and a contemporary Tokyo appearance. She has a modern dark layered hairstyle with natural movement and face-framing pieces. Her clothing has a strong current Y2K streetwear influence, but it must look fashionable and authentic rather than like a Y2K costume. Use fitted layers, subtle metallic details, a compact shoulder bag, and contemporary footwear. She should look like someone you could genuinely encounter in a fashionable Tokyo neighborhood today. Keep Woman 1's face, body, hairstyle, clothing, and accessories consistent throughout the entire video. Her skin must have realistic natural texture and visible detail. Avoid glossy skin, plastic-looking faces, CGI appearance, or an animated AI-video look. SCENE 1 — approximately 4 seconds Woman 1 walks along an active Tokyo street at night and enters a contemporary convenience store. Keep Woman 1 clearly visible while establishing the environment. The street feels active and lived-in, with pedestrians, traffic, storefront illumination, and authentic Japanese signage. Her movement into the store should feel spontaneous and natural rather than staged. SCENE 2 — approximately 7 seconds Inside the convenience store, Woman 1 selects a cold drink from a refrigerated case and walks to a modern self-checkout terminal. The store should look like an ordinary, contemporary Tokyo convenience store, not a futuristic interpretation: bright practical lighting, organized narrow aisles, refrigerated cases, shelves filled with believable Japanese products, and modern checkout technology. At the self-checkout area, Woman 1 notices an interesting local event flyer posted nearby. She pauses naturally, looks at it with genuine interest, takes out her phone, and photographs it. This is a small incidental discovery, not a dramatic moment. She then scans her drink and completes the purchase using contactless payment. SCENE 3 — approximately 4 seconds Woman 1 puts her phone away, takes her drink, and exits through the automatic doors. She steps back onto the active Tokyo night street and naturally joins the pedestrian flow. End with her continuing through the environment. She should not stop and pose for the camera. DRINK AND PACKAGING The drink is a realistic 500 mL PET bottle with an original fictional Japanese beverage brand. Give it professionally designed contemporary Japanese packaging with believable label hierarchy, restrained typography, realistic product information, and proper Japanese writing. Do not reproduce an existing beverage brand. The product should simply look convincing enough that it could genuinely be sitting in a Tokyo convenience-store refrigerator. TEXT AND ENVIRONMENTAL AUTHENTICITY Any visible Japanese signage, product packaging, event-flyer typography, or other readable environmental text must consist of real, correctly written Japanese or contextually appropriate English. Do not generate gibberish, fake Japanese characters, meaningless words, or obviously distorted AI typography. VISUAL STYLE Prioritize photorealism above everything else. The video should have believable human movement, realistic physics, natural skin texture, ordinary environmental imperfections, and authentic nighttime lighting. It should feel like something captured naturally on a high-end smartphone while following Woman 1 through Tokyo, not like a commercial, fashion shoot, CGI render, or highly polished cinematic production. Maintain the quality and realism of the source imagery throughout the video. Do not introduce an animated or synthetic appearance as the video progresses. COST CONTROL This is also a credit-usage test. Use the minimum number of generation operations necessary to produce this one finished 15-second video. Do not automatically add previews, prompt enhancement, additional generations, alternate versions, upscaling, or other credit-consuming operations. Before generating, tell me whether MiniMax H3 Max can actually produce 3:4 at 552 × 736. If it cannot output those exact dimensions, tell me the closest supported resolution and aspect ratio and wait for my approval before generating anything. Also tell me the estimated total credit cost, if available, before starting the generation. Do not generate anything until I explicitly approve the production after you answer those two questions.

關於 MiniMax H3

MiniMax H3 是 MiniMax 海螺视频系列的旗舰,最大的特点是输入端真正多模态、输出端一步到位。你可以从一段文字描述、一张参考图、一段已有视频、一条音轨或它们的组合开始生成——而结果自带与画面同步生成的原生立体声,不是事后配上去的。

这个组合让它成为「成片感」需求的首选:靠环境声带货的广告片、靠脚步声和空间音撑住情绪的叙事镜头、需要让静态照片带着可信的运动和声音活起来的图生视频。它继承了海螺系列著称的物理真实感,对推轨、环绕、手持等镜头指令的服从度也格外高。

用过海螺 02 的人会觉得 H3 很熟悉,但它对「含糊」远没那么宽容。老一代给一句场景描述就能替你把剩下的补齐;H3 摆出六七个彼此独立的控制杆——主体、动作、场景、镜头、光线、声音,以及输入模态本身——你不拉的那根,它就按自己的默认值还给你。控制力更强这件事,只有你真的花掉它才算优势。

由此带来的实际差别是:一条 MiniMax H3 提示词更像分镜表,而不是图注。写「雨中行走的女人,电影感」,你会拿到一条合格但没有面孔的片子;写清镜头焦段、她行走的方向、摄影机相对她的运动、光的质感、雨落在什么材质上是什么声音——你拿到的才是一条真能剪进片子里的镜头。

正因为 H3 接受这么多种输入,提示词就是告诉它「这次做哪种活」的唯一途径——写得含糊,恰恰浪费了它最强的可控性。上方画廊收录了经过验证的 MiniMax H3 提示词及真实生成视频,直接复制一个已跑通的结构,换上你自己的主体、运动和声景即可。

本页所有提示词都可以免费复制:不用注册、没有付费墙、提示词正文没有水印。拿走结构,换掉名词,在你能用到模型的地方跑就行。

如何寫好 MiniMax H3 提示詞

  1. 1按「主体 → 动作 → 场景 → 镜头 → 光线氛围 → 声音」的结构写。H3 六层都读;漏写哪层,哪层就由模型替你决定。
  2. 2把它当镜头写,不要当图注写。「电影感」「高质量」是最不值钱的两个词——换成你真正想要的那件具体的事:焦段、调色、时辰、天气。
  3. 3用真实的镜头语言:「缓慢推轨」「手持跟拍」「绕主体环绕」。H3 对有名字的镜头运动的执行精度,远高于「动感镜头」这类空泛说法。
  4. 4要慢动作,就把它写成画面的属性、而不是人物的属性:「以慢动作拍摄,水珠悬停在空中」有效;「她走得很慢」只会让她以正常速度慢吞吞地走。想在片中变速,就写清变在哪一刻——「实时速度,玻璃落地瞬间转入慢动作」。
  5. 5把声音明确写出来——声音是原生生成的,它真的听:「铁皮屋顶的雨声、远处闷雷、不要配乐」或「明快合成器配乐在旁白下渐强」。不写声音,得到的就是泛泛的环境音。
  6. 6写材质,不要只写声音。「脚步声」太泛;「靴子踩在湿砾石上」才给了模型可以合成的材料。从「雨声」到「引擎声」,这个办法一律有效。
  7. 7如果你要的是没有声音,那就明说——「无配乐、无对白,只保留室内环境音」。安静也是一条 H3 会照办的指令,前提是你下了这条指令。
  8. 8从参考图出发时,只描述运动和镜头——图片已经定了主体、风格和构图,再复述一遍反而会引入漂移。
  9. 9每次生成只给一个连续的动作节拍。一条片子想塞三个故事时刻,三个都会糊掉;故事更长就分段生成再剪。
  10. 10把你的画面基调冻结成一条可复用的子句——胶片质感、调色风格、镜头特性——整个 campaign 保持一字不改,成片才能剪到一起。
  11. 11要跨镜头保住同一个人物,就把同一张参考图喂进每一次生成,而不是反复用文字描述这个人。文字描述的脸会在多次生成之间漂移,图片不会。
  12. 12结果没打中时,只改一层再生成,别整条重写。整条重写你就再也说不清到底是哪一句起的作用。

常見問題

MiniMax H3 的提示词该怎么写?

按分镜表的顺序写:主体、动作、场景、镜头运动、光线氛围、声音。H3 把这几层各自当成一条独立指令,你留白的那一层,它就用自己的默认值填。每次生成只放一个连续动作,镜头运动用真实的摄影术语,声音一定要点名。上方画廊里的每一条提示词都是这个形状——最快的学法就是复制一条,把主体换成你自己的。

MiniMax H3 是免费的吗?

本页的提示词是免费的:无需账号、没有付费墙,随便复制。模型本身要通过 MiniMax 的海螺 AI 或 MiniMax 开放平台 API 使用,具体的套餐和免费额度由 MiniMax 那边定,当下怎么算请以他们的页面为准。

MiniMax H3 最擅长什么?

需要「拿到即成片」的片段:画面与立体声一次生成,输入端不挑食——文字、静态图、已有片段、音轨都行。物理真实感和精准的镜头服从度,让它在广告、电影感长镜头和图生视频场景里格外强。

MiniMax H3 会生成声音吗?

会——立体声与画面原生同步生成,不是后期配音。要发挥它,就在提示词里直接导演混音:写清环境声、音效、要不要配乐。什么都不写,得到的就是泛泛的环境音。

怎么让 MiniMax H3 生成慢动作?

把慢动作描述成画面的属性,不是人物的属性。「以慢动作拍摄」或「高速摄影机拍摄、慢速回放」改变的是这条片子怎么被拍下来;「她动作很慢」只会让主体在正常速度里显得迟钝。再配一个只有慢下来才看得见的细节——水花、尘土、布料、发丝——让效果有东西可展示;想中途变速,就点名变在哪一刻。

可以不用文字,从图片或音轨开始吗?

可以。H3 输入端是多模态的:参考图可以定主体和构图,已有片段可以延展或改风格,音轨可以驱动整个场景。当图片已经承载了画面风格时,文字提示词只管运动和镜头就好。

MiniMax H3 和海螺 02 有什么区别?

它们是同一条海螺视频线的两代。海螺 02 给一句场景描述,剩下的决定大多替你做了;H3 输入端多模态、立体声与画面原生同生,并且把更多决定权当成可控项交还给你——这意味着它奖励写得细的提示词,也比老一代更不容忍偷懒的提示词。

为什么 MiniMax H3 漏掉了我提示词里的一部分?

通常是因为一条提示词同时要了好几件事。H3 处理单个连续动作节拍很稳,一条覆盖三个故事时刻的提示词会被平均成一团糊。另一个常见原因是把参考图已经定死的东西又用文字描述了一遍——图说一套、文字说另一套,结果就漂。把提示词砍到一个动作节拍,一次只改一层,再生成。

在哪里可以试用这些 MiniMax H3 提示词?

H3 可在 MiniMax 的海螺 AI 中使用,开发者可走 MiniMax 开放平台 API。从本页复制任意提示词粘贴过去,换成你自己的主体、镜头运动或声景即可。