精选可直接复制的 MiniMax H3(海螺 3)视频提示词,每条都附示例片段与作者署名。悬停播放,一键复制。
最近更新: · MiniMax H3 模型与价格 · API 文档
MiniMax H3(社区称海螺 3.0)是 MiniMax 于 2026 年 7 月 31 日发布的多模态视频模型:2K、24fps、4-15 秒,自带同步原生音效与对白,支持文生视频、首尾帧控制、参考图/参考视频、动作迁移和生成式视频编辑,单次最多 9 张参考图 + 3 段参考视频 + 3 段参考音频,官方 API 约 $0.13/秒起。本页收录可直接复制的 H3 提示词,每条附示例片段与作者署名,免费使用。
出处:MiniMax(模型方官方站点)
Style: <image_1> (don't reproduce the image, use it only as aesthetic reference) Edit: Comedy, fast paced, mixing close ups, medium shots and wide shots from uncanny and tilted angles. Super-imposed texts with doodles identify ingredients as they appear and add comic remarks. Scene: In Porto, Portugal a crazy chef explains how to do a Francesinha in Portuguese from Portugal with Porto accent.
Use the supplied image as the exact opening frame. Create a hilarious, high-budget 1980s live-action family comedy movie scene, photographed on a real soundstage with practical robot costumes, animatronics, handmade props, vintage wardrobe, and authentic 35mm film color. The entire group suddenly hears something above them. Everyone’s eyes dart upward at the same moment. The men, women, children, and robots dramatically tilt their heads back, point toward the ceiling, and erupt into chaotic celebration. They gasp, scream, laugh, jump, wave their arms, slap each other on the shoulders, and completely lose their minds with exaggerated 1980s comedy reactions. The children bounce excitedly. The adults stumble into one another. The robots flash their eyes, flap their mechanical arms, spin clumsily, and celebrate with funny practical animatronic movements. One man shouts: “OPEN SOURCE?!” The entire crowd answers together: “MINIMAX H3!” They cheer wildly as one robot attempts a victory dance, loses its balance, and is caught by the shocked family at the final second. Fast, energetic comic timing with believable ensemble choreography. Begin with a brief locked group shot, then a subtle handheld push-in as the chaos escalates. Keep every original person, robot, shirt design, and the large headline clearly recognizable. Preserve facial identity, wardrobe, composition, lighting, and 1980s production design. Authentic optical softness, warm tungsten lighting, rich film grain, slight gate weave, practical effects, natural motion blur, expressive physical acting, polished studio-comedy sound mix, triumphant synthesizer sting, cheering, robot beeps, and comedic percussion. No CGI, no digital-looking robots, no morphing, no extra people, no duplicated characters, no distorted faces, no rewritten text, no spelling changes, no modern clothing, no style shifts, and do not turn the scene into animation.
Grounded photorealistic live-action relationship drama, natural 24fps motion. FORMAT Two connected 15-second clips, 16:9, grounded photorealistic live-action relationship drama, natural 24fps motion. SHARED STYLE A small living room at night. One warm floor lamp mixes with cold blue streetlight through the curtains. A framed wedding photograph hangs behind the couple. Restrained handheld camera, realistic skin and fabric, tense silence, no melodramatic shouting. CONTINUITY The husband remains frame left beside the coffee table, holding his wife’s unlocked phone containing the evidence. The wife remains frame right near the window. Preserve their faces, clothes, positions and screen direction across every cut. CLIP 1 — THE CONFESSION 0.0–5.0 Camera: Slow 35mm push toward both characters. Action: The husband stares at the phone, then raises his eyes toward her. HUSBAND: “How long has this been going on?” 5.0–8.0 Camera: Cut to a restrained 50mm close-up of the wife. Action: She cannot meet his eyes. After a short pause: WIFE: “Six months.” 8.0–15.0 Camera: Close on the husband. His grip tightens around the phone, but he controls his voice. HUSBAND: “Six months? I gave you twelve years. Why?” Final frame: The wife finally looks directly at him, preparing to answer. CLIP 2 — THE REASON Begin from the exact positions and expressions inherited from Clip 1. 0.0–6.5 Camera: Slow push toward the wife as she folds her arms defensively. WIFE: “Because I’m tired of being broke. Every dream with you becomes ‘maybe one day.’” 6.5–10.0 Camera: Cut to the husband. His anger gives way to disbelief. HUSBAND: “So you chose him for his money?” 10.0–15.0 Camera: Return to the wife. She holds his gaze without apologizing. WIFE: “I chose someone who can give me the life I want.” Final frame: Hold on the husband as the meaning lands. His hand lowers, and the phone slips onto the sofa.
Two characters arguing in English. Make the emotion feel very realistic angry, tense, the friction between them should be palpable. Shaky handheld camera. No music.
SCENE CONTEXT A middle-aged man stands in the middle of a dirt road in open desert and holds a pistol level at the person filming him. He gives an instruction, is answered by name, and warns them not to say it again. Nothing is fired. TIMELINE 0.0–3.0s — he is already aimed and steady; his first line lands. 3.0–6.0s — the off-frame reply uses his name; his jaw sets and the muzzle does not drop. 6.0–9.0s — his second line; he takes half a step forward and re-sights. 9.0–11.0s — the off-frame voice starts his name again and stops. 11.0–15.0s — his last line; wind lifts dust across the road behind him; he holds the aim. Hold to black-free end. ACTIVE REFERENCES No image or video references are active in this shot. Build the man from description only. Man, fifties, white, average build, standing, thinning brown hair, a brown moustache, thin-rimmed prescription glasses; light green long-sleeved button-down shirt with the sleeves buttoned at the wrist, dark work trousers; stern, brow furrowed, sweat starting at the hairline; frightened underneath and controlling it; speaking voice flat, plain, mid-western American, no theatricality. Second speaker: the unseen person the pistol is aimed at, occupying the camera position, never entering frame and never seen. Voice only — male, forties, close to the lens, low and careful. LOCATION MAP Camera on the dirt road 3.5 m from him at his chest height, facing him square-on so the muzzle is aligned with the lens axis. FG: rutted dirt and loose gravel across the lower frame. MG: the man alone, centre frame. BG: the dirt road receding to a vanishing point at x 50%, y 44%, low scrub and dry brush from x 0% to x 30% and x 70% to x 100%, a distant fence line at x 62%, y 48%, flat desert horizon at y 46%, overcast sky filling the upper half with one pale blue break low at x 24%, y 36%. He occupies x 50%, y 50%, body frontal, both feet planted, right arm extended toward camera with the silver pistol at x 50%, y 56%, held at the very front of the depth plane. His eyeline is straight down the lens. The first visible frame already has him aimed and in position — no empty establishing frame, no walk-in, no delayed reveal, and he remains the only figure in frame for the entire take. FORMAT MODE Single continuous take, no internal cuts, no transitions, no fades. Real-time motion. His lines are on camera; the second voice is an intentional off-frame speaker at the camera position. No other voices, no narration, no subtitles, no score. CAMERA 47° diagonal field of view, camera 3.5 m from him at chest height, natural human-eye perspective, no distortion, natural face and body proportions, the road and horizon readable behind him, comfortable depth of field with the pistol slightly forward of his focal plane. Handheld, shoulder-mounted mass, operator breath and small involuntary settling — the camera is being held by the person he is aiming at, so it flinches minutely rather than gliding. Total travel under 20 cm, no push-in, no zoom, no focus rack. Focus holds his eyes behind the glasses; the pistol stays just soft. ACTION TIMING He is already sighted along the barrel as the shot opens and speaks without lowering it. The off-frame voice answers with his name; his jaw sets and the muzzle does not waver. He speaks his second line and takes half a step forward on the dirt, re-sighting as his weight lands. The off-frame voice begins his name again and stops. He delivers his last line, and wind lifts a curtain of dust across the road behind him. He holds the aim to the end without firing and without lowering the weapon. PHYSICS The pistol carries real weight — his wrist and forearm show the load and micro-correct continuously; the muzzle drifts a few millimetres and is pulled back on line. His half step lands heel-first in soft dirt with visible weight transfer and a small dust puff that drifts and thins. Shirt fabric pulls across the extended shoulder and creases at the elbow. Wind moves the loose dust and the dry scrub in the same direction and at the same time. Thinning hair lifts and settles with the gust. Sweat beads at his temple and does not run. LIGHTING Primary source is the full overcast sky acting as one enormous soft toplight, giving even illumination with minimal harsh shadow — a soft shadow under the brow and the nose only, no hard edges anywhere. Secondary is dry ground bounce filling under the chin and lifting the shadow side of the pistol. Both lenses of his glasses carry a broad soft reflection of the sky, and his eyes stay readable behind them. Exposure is set for his face; the sky sits bright but not clipped. No sun shaft, no lens flare, no added key, no light change during the take. AUDIO Dry wind across open ground, loose grit moving on the road, dry brush ticking, one distant bird, his own breathing close and unsteady. Ambient ducks under the voices. Exactly this exchange, flat and plain, no shouting — 25 words total: HIM: "Put it on the ground. Slowly." OFF-FRAME: "You're not going to shoot me, Ray." HIM: "I've been wrong about myself before." OFF-FRAME: "Ray—" HIM: "Don't say my name again." His first line opens the shot. The off-frame reply lands as his jaw sets. His second line comes as he takes the half step. The off-frame "Ray—" is cut off by his last line, which is quieter than everything before it. His lips move only for his own lines; there is no dialogue in the pauses and no muttering between beats. POSITIVE LOCKS Exactly one visible person is in frame for the whole take — nobody enters at any edge, no second figure appears on the road, at the fence line or in the reflections on his glasses, and the off-frame speaker is never seen. The pistol stays in his right hand, stays aimed at the lens, and is never fired, never lowered and never re-holstered; no muzzle flash, no recoil, no gunshot. He stays on the road and never turns away from camera. Naturalistic muted palette of dry greens and browns against pale overcast grey. Live-action photoreal footage shot on ARRI Alexa 35, tense neutral mood, natural film grain.
16:9, 15s, hyper-realistic Korean noir crime teaser. A rain-soaked Korean woman (late 20s, trench coat) enters an abandoned underground nightclub searching for her missing sister. A scarred crime boss (40s), presumed dead for years, emerges from the shadows with a lit cigarette. Atmosphere: 90s Korean noir, practical neon, rain, smoke, film grain, tactile realism, no CGI gloss. Shots: 1. She enters: "언니... 여기 있어?" 2. Lighter ignites, revealing his scar. "오랜만이네." 3. She freezes: "당신... 죽은 줄 알았는데." 4. Slow push-in. He smiles: "네 언니가 날 죽였다고 생각했겠지." 5. They lock eyes. Thunder. Cut to black. Audio: Rain, thunder, jazz crackle, lighter click, intimate silence during dialogue.
Duration: 15 seconds | Aspect ratio: 16:9 | Style: Cinematic drama, emotional cross-cutting dialogue SCENE A distressed woman with an updo wearing a beige turtleneck sweater and holding a dark coat in an airport terminal gate at night. A somber man in a grey t-shirt inside a dimly lit kitchen with rain-speckled windows. SHOT BREAKDOWN 0-4s — Medium shot of the woman standing among rows of empty blue airport seats, speaking anxiously on her phone. 4-8s — Cut to a medium shot of the man leaning against a kitchen counter, listening solemnly on his phone. 8-12s — Cut to a tight close-up of the woman, teary-eyed and speaking with deep emotion. 12-15s — Cut to a tight close-up of the man closing his eyes with a pained, heartbroken expression. CAMERA 50mm and 85mm prime lens feel, shallow depth of field, static framing, eye-level medium shots cutting cleanly to intimate close-ups. LIGHTING & PALETTE Airport setting uses cool white overhead lighting, dark glass reflections, deep blues, and beige. Kitchen setting features warm pendant lighting, heavy shadows, and amber outdoor bokeh. AUDIO Low airport terminal hum, soft ambient drone, distant rain against window glass, quiet emotional phone dialogue. AVOID Camera shake, daylight, fast whip pans, exaggerated body motion, upbeat background music.
提示词公式
主体 + 具体动作 + 镜头运动 + 光线与色调 + 环境细节 + 声音/台词 + 速度节奏 + 画幅
例:雨夜的居酒屋门口,一位穿米色风衣的女人收伞回头,镜头从她肩后缓缓推近到侧脸特写;冷蓝街灯混着暖橙灯笼光,雨滴打在伞面;背景有雨声与远处电车驶过的轰鸣,她轻声说「你来晚了」;实时速度,16:9。
MiniMax H3(社区常称海螺 3.0)是 MiniMax 于 2026 年 7 月 31 日发布的多模态视频模型:输出 2K、24fps、4-15 秒的视频,并能直接生成同步的原生音效与对白;支持文生视频、首尾帧控制、参考图/参考视频、动作迁移与生成式视频编辑,单次最多可喂 9 张参考图、3 段参考视频、3 段参考音频。
免费。每条都可以直接复制走,拿到任何能调 H3 的地方去跑,不需要登录也不需要付费。
每张卡片下方都标了提示词作者,点署名可以跳到原帖/主页。我们只做收录与整理,版权归原作者;如果你是作者、希望撤下或更正署名,发邮件给我们即可。
H3 已经在 apimodels.app 上线,模型名 minimax-h3:原生 2K + 同步音频,时长 5-15 秒任意整秒,单次最多 9 张参考图 + 3 段参考视频 + 3 段参考音频。按秒计费 $0.145/秒(5 秒 $0.725),参考图前 5 张免费、第 6 张起每张 $0.03,只对成功请求计费。同一个 key、同一套端点还能调 Seedance、Kling、VEO、Grok。此外你也可以在 MiniMax 官方 API 或海螺(Hailuo)产品里跑 H3。
把一条提示词当成一张分镜卡来写:主体在做什么、镜头怎么动、光线和色调、环境细节,再加上你想要的声音(H3 会生成原生音效和对白)。时长控制在 4-15 秒能覆盖的范围内——一条提示词只讲一个动作,比塞三段剧情效果好得多。
按 MiniMax 的使用条款执行,请以官方条款为准。本页收录的提示词文本你可以自由修改、二次创作。
MiniMax H3 已经上线,和另外 85+ 个图像、视频、语音、大模型共用一个 API、一个 key,按量计费、注册送 $1。
我们反推的提示词以 MIT 授权全文开放,作者原文则带署名与原帖链接索引。欢迎提 issue 补充新的提示词与成片。
在 GitHub 上查看