MiniMax H3 · 海螺3 · Hailuo H3 提示词库

MiniMax H3 提示词免费海螺3提示词库——真实视频,附完整提示词

下面每条提示词旁边就是它实际生成的视频,所以你复制的每条 MiniMax H3 提示词都是已经跑通过的。再用海螺3提示词指南控制主体、动作、运镜、节奏与原生音频。MiniMax-H3、海螺3、Hailuo 3、Hailuo H3 指的是同一个视频模型,本页所有提示词模板对这些名称都通用。

文生视频提示词图生视频提示词参考生视频提示词原生音频与对白来源配对案例创作者署名免费复制

MiniMax H3 提示词案例——观看、复制、核验来源

真实 MiniMax H3 视频与海螺3原始提示词

按用途或生成模式(文生视频、图生视频、参考生视频)筛选这些海螺3提示词案例。每张卡片都可播放原视频,并包含完整公开提示词、作者署名、X 原帖及核验备注——你复制的每条 MiniMax H3 提示词,上方就是它生成出来的成片。

分类:
生成模式:
精选

山谷回声与轴线后拉镜头

真人登山者向山谷呼喊,通过身份锁定、沿光轴后拉、距离衰减与左右山体回声控制画面和声音。

图生视频电影感16:9请求时长: 9s
使用此提示词
查看完整原始提示词
integrated_multimodal_description: [Shot 1] Live-action, cinematic and photorealistic. The young female hiker shown in <Picture 1> remains on the narrow mountain ridge, preserving her exact facial identity, natural skin texture, brown hair, red insulated hiking jacket, backpack straps, body proportions, front-facing orientation, lighting, background mountains and initial composition.

The camera begins in the close head-and-shoulders framing established by <Picture 1>. She remains unaware of the camera and keeps her eyes fixed on one distant point in the valley ahead. A light mountain wind moves only a few loose strands of her hair and slightly shifts the fabric around her collar.

She takes one natural breath, gently raises her chin and loudly calls into the valley. The young woman with a clear, strong, natural American voice (S1) shouts with joyful energy: <d>[English] Helloooooooo!</d>

Her lips, jaw, cheeks, throat and chest move naturally with the sustained call. Her arms remain relaxed beside her body. Her feet, hips, shoulders and head remain oriented in the same direction; she does not rotate, step, turn, or follow the camera with her eyes.

As she begins shouting, the camera pulls out with large amplitude at slow speed while pedestaling upward with small amplitude at slow speed. It travels directly backward along the same optical axis, progressing continuously from a close portrait to a chest-up frame, then to a full-body view, and finally to a wide view of the ridge and surrounding valley.

The camera does not orbit, truck sideways, pan around her, or use a digital zoom. Realistic parallax reveals the steep grassy slopes, distant layered mountains and deep valley while the horizon remains stable.

As the camera moves farther away, her direct voice becomes quieter and more distant. After a short natural delay, the final syllable reflects first from the left mountain wall and then more softly from the right side of the valley, creating separated stereo echoes that gradually become quieter and more diffuse.

She completes the call and remains standing in exactly the same place and orientation. By the end of the shot, she appears as a small red figure on the vast green ridge while the final distant echo fades naturally across the valley.
精选

APEX RIDGE 赛车选择界面

使用四张车辆参考图,在精确时间轴内切换赛车,并保持游戏 UI、车型和最终选中状态稳定。

参考生视频游戏与 UI16:9请求时长: 6s
使用此提示词
查看完整原始提示词
Create a polished 6-second, 16:9 photorealistic racing-game car-selection sequence for the fictional game APEX RIDGE.

REFERENCE MAP
[REF_1]: VULCAN S1 — crimson red track-focused supercar.
[REF_2]: TITAN RS — matte silver high-performance GT.
[REF_3]: KAIRO X — yellow electric hypercar.
[REF_4]: complete APEX RIDGE car-selection interface, visual-layout reference and final metallic-blue AURELIS R9 appearance.

GLOBAL CONTINUITY
Recreate the premium dark showroom and game UI from [REF_4]. Keep the exact title “APEX RIDGE” and subtitle “SELECT YOUR CAR” visible and perfectly stable throughout. Preserve the same camera position, interface layout, showroom lighting, vehicle scale and perspective. Large white left and right navigation chevrons remain visible beside the active vehicle. The bottom carousel always contains exactly four vehicle cards in this order: VULCAN S1, TITAN RS, KAIRO X, AURELIS R9. The specifications panel always uses the labels SPEED, ACCELERATION, BRAKING and HANDLING. Use clean, readable English UI text only. No real-world brands, badges or logos.

TIMELINE — EXACTLY 6.0 SECONDS

0.0–1.5 seconds — VULCAN S1
Open on the red VULCAN S1 from [REF_1] as the large central vehicle. Its bottom card gains a sharp red-white selection outline. The right chevron gives a subtle luminous pulse. The four specification bars animate rapidly to their VULCAN S1 values. Add a restrained turntable movement of only a few degrees and realistic moving reflections across the red bodywork.

1.5–3.0 seconds — TITAN RS
At exactly 1.5 seconds, a crisp right-arrow click triggers a fast horizontal carousel slide with elegant motion blur. The red vehicle exits left and the silver TITAN RS from [REF_2] enters from the right, settling in the identical central position. The TITAN RS card becomes highlighted, the previous card returns to neutral, and all four specification bars smoothly update. Preserve body design and silver color exactly.

3.0–4.5 seconds — KAIRO X
At exactly 3.0 seconds, the right chevron pulses and clicks again. Perform the same quick premium carousel transition. The silver GT exits left and the yellow KAIRO X from [REF_3] enters from the right. Highlight only the KAIRO X card. Animate the stats to new values and sweep a thin light reflection across the yellow body without changing its shape.

4.5–6.0 seconds — AURELIS R9 FINAL SELECTION
At exactly 4.5 seconds, trigger the final right-arrow click. The yellow car exits left and the metallic-blue AURELIS R9 shown in [REF_4] enters from the right and locks perfectly into the hero position and design shown in [REF_4]. Highlight the AURELIS R9 card with a vivid electric-blue border, reveal the exact label “SELECTED”, and animate the final stats to SPEED 92, ACCELERATION 95, BRAKING 88, HANDLING 91. The blue CONFIRM button illuminates and gives one controlled pulse. Finish on a clean, stable hero frame of the selected blue car with the full interface readable.

MOTION AND CAMERA
Locked front three-quarter showcase camera; only subtle showroom parallax and small turntable movement. Carousel transitions are fast, precise and game-like, never cinematic scene cuts. Every car remains sharply recognizable and consistent with its corresponding reference image. No camera shake, no zoom jump, no morphing and no change of environment.

AUDIO
Premium racing-game UI sound design: one soft navigation click and short aerodynamic whoosh at 1.5, 3.0 and 4.5 seconds; a distinct half-second engine-preview sound for each car; finish with a clean electronic confirmation tone when AURELIS R9 becomes SELECTED. Keep audio tight and synchronized. No dialogue and no voice-over.

NEGATIVE CONSTRAINTS
No extra cars, no missing cars, no duplicated vehicles, no malformed wheels, no model morphing, no color changes, no illegible text, no misspelled labels, no flickering UI, no random icons, no real manufacturer marks, no racing footage and no transition away from the selection showroom.
精选

APEX RIDGE 下山遥测赛段

多机位下山赛段,要求遥测 HUD 持续可读、道路空间连续、车辆物理和声音同步。

参考生视频游戏与 UI16:9请求时长: 15s
使用此提示词
查看完整原始提示词
TITLE: APEX RIDGE - DOWNHILL TELEMETRY RUN

FORMAT
Exactly 15 seconds, 16:9 widescreen, ultra-photorealistic AAA racing-simulator gameplay with native synchronized stereo audio. Fast, premium motorsport presentation with multiple camera viewpoints, physically credible vehicle dynamics and strict spatial continuity. The entire sequence takes place on a closed competition road with no public traffic.

OMNI REFERENCES
- Use [REF_1] as the exact exterior vehicle reference. Preserve the same fictional cobalt-blue supercar body, proportions, headlights, carbon details, black wheels, copper brake calipers and rear wing in every exterior view.
- Use [REF_2] as the exact cockpit reference. Preserve the steering wheel, carbon dashboard, copper stitching, paddle shifters, windshield geometry and black racing gloves.
- Use [REF_3] as the exact mountain-road, lighting and downhill-action reference. The car must always travel from the high mountain pass DOWN toward the valley through successively lower hairpins. Never depict uphill travel.

PERSISTENT RACING HUD - VISIBLE FOR ALL 15 SECONDS
Keep one stable, clean, semi-transparent racing-simulator telemetry overlay visible in every shot, including cockpit, chase, wheel and aerial views. The HUD remains anchored to the screen safe area and never disappears during camera cuts.
- Lower right: large digital SPEED in km/h, current GEAR, curved RPM tachometer and redline.
- Lower left: horizontal BRAKE and THROTTLE input bars, with brake in red and throttle in green.
- Upper left: compact downhill route map with a moving position marker and the next hairpin highlighted.
- Upper right: running sector timer and delta.
- Small central lower strip: steering input, ABS and traction-control activity.
Telemetry must update logically and continuously with the driving: speed falls under braking and rises on exit; RPM drops with each downshift and climbs under acceleration; gear changes sequentially; brake and throttle never activate fully at the same time. Keep all indicators crisp, readable, correctly spelled and visually consistent. No flicker, random values, duplicated widgets or HUD redesign between shots.

DRIVING AND CAMERA TIMELINE
0.0-3.5 seconds - COCKPIT VIEW
Begin directly inside the cockpit from [REF_2], descending rapidly toward a tight right-hand hairpin visible below. Speed reads approximately 178 km/h in 5th gear at high RPM. The driver brakes firmly; the red BRAKE bar rises, the car's nose loads forward naturally, and the speed decreases smoothly. Two synchronized paddle downshifts: 5th to 4th, then 4th to 3rd, each with a realistic rev-matched engine blip. Gloved hands rotate the wheel progressively right. No camera shake beyond subtle chassis vibration.

3.5-7.0 seconds - REAR THREE-QUARTER CHASE VIEW
Cut to a low rear chase camera that preserves the exact car from [REF_1]. The car enters the downhill right-hand hairpin in 3rd gear at approximately 92 km/h. Brake lights glow during trail braking, the outside suspension compresses, tires follow the asphalt without sliding, and the car clips the inside apex cleanly. The road and valley remain visibly lower ahead. The HUD stays fixed on screen and continues updating through the cut.

7.0-10.0 seconds - FRONT-WHEEL / FENDER VIEW
Cut to a stabilized camera mounted just behind the front-left wheel, showing the spinning black wheel, copper brake caliper, damp asphalt and barrier rushing past. The driver releases the brake; the red bar falls to zero and the green THROTTLE bar rises progressively. Speed increases from roughly 96 to 128 km/h. Shift cleanly from 3rd to 4th as the RPM reaches the shift point. Show realistic tire deformation, suspension recovery and restrained motion blur while the car body remains stable.

10.0-13.0 seconds - HIGH AERIAL TRACKING VIEW
Cut to a high cinematic tracking shot revealing the car descending from the completed hairpin toward the next lower switchback, matching [REF_3]. The camera moves parallel to the slope while preserving screen direction and exact road geography. The cobalt-blue car accelerates through the short downhill straight, with the valley far below and several lower curves visible ahead. Telemetry remains readable and correctly synchronized.

13.0-15.0 seconds - HOOD VIEW AND SECTOR RESULT
Cut to a low hood-mounted forward view. The road drops sharply toward the next left-hand hairpin. The car reaches approximately 151 km/h in 4th gear, then the driver begins braking at the final half-second. Show a brief, tasteful HUD notification: “SECTOR 2 - PERSONAL BEST” while all core telemetry remains visible. End in motion with the engine note falling under braking, ready to continue into another clip.

VEHICLE PHYSICS
Authentic high-performance simulation: correct downhill acceleration, progressive braking distance, forward weight transfer, lateral suspension load, stable tire contact patches, sequential paddle shifts, realistic steering angle, physically coherent wheel rotation and consistent road grip. The driver follows the racing line and remains fully in control. No drifting, jumping, collision or arcade behavior.

AUDIO
Native synchronized stereo game audio with no music and no narration. Detailed naturally aspirated high-performance engine, intake resonance, exhaust reflections from rock walls, rev-matched downshift blips, paddle clicks, transmission whine, tire scrub under load, brake hiss, suspension movement, gravel ticks beneath the chassis, wind pressure and subtle cockpit vibration. Audio perspective changes naturally with each camera: enclosed and detailed inside the cockpit, louder exhaust in chase view, mechanical tire and brake detail at the wheel, wider mountain echo in the aerial view. No audio cuts, distortion or artificial looping.

VISUAL STYLE
Ultra-realistic premium racing game capture, physically based materials, ray-traced reflections, realistic damp asphalt, natural cold sunrise, thin alpine mist, restrained cinematic contrast, believable motion blur, sharp vehicle identity and stable mountain geography. Smooth professional cuts with no teleportation.

NEGATIVE CONSTRAINTS
No real car manufacturer, existing racing-game branding, logos, badges, license-plate text, watermark, subtitles, public traffic, spectators, pedestrians, police, road debris, crash, collision, drifting, uphill motion, reversed travel direction, changing car design or color, changing cockpit, extra hands, malformed fingers, steering-wheel deformation, impossible wheel rotation, floating tires, incorrect reflections, warped barriers, discontinuous road, impossible camera positions, HUD disappearance, HUD flicker, unreadable telemetry, misspelled labels, random speed jumps, impossible gear changes, simultaneous full brake and throttle, duplicated UI, arcade effects, nitrous flames or music.
精选

第一视角厨房高压清洗变换

通过高压水流物理、第一视角双手和厨房参考图完成由脏到净的连续变换。

图生视频变换效果9:16请求时长: 10s
使用此提示词
查看完整原始提示词
Use **@Image1** as the exact visual reference for the finished kitchen. Preserve the same room layout, island, cabinets, appliances, pendant lights, stools, countertops, flooring, lighting, materials, and camera composition.

Create a **10-second ultra-realistic POV pressure-washing transformation video** in **vertical 9:16**.

## Camera

First-person perspective from a worker holding a powerful pressure-washer wand. Two realistic hands remain visible throughout. Keep the kitchen geometry and camera direction consistent with **@Image1**.

Use energetic handheld movement, realistic recoil from the water pressure, fast sweeping motions, close-up spray passes, water droplets on the lens, mist, splashes, and subtle motion blur.

## Scene 1 — 0–2 seconds

The entire kitchen is completely covered in thick black dirt, soot, grease, mud, and grime. The cabinets, island, countertops, appliances, stools, walls, pendant lights, and floor appear almost entirely black.

The worker raises the pressure-washer wand and activates it. A powerful jet of water blasts the front edge of the island, immediately cutting a clean bright line through the dirt.

## Scene 2 — 2–7 seconds

The worker rapidly pressure-washes across the kitchen in continuous sweeping movements.

Black grime peels away in satisfying strips from the island, cabinets, refrigerator, countertops, backsplash, cooker hood, stools, and floor. Dirty water flows downward and spreads across the floor while mist and dark particles fill the air.

Use speed ramps and quick whip movements between surfaces while every cleaned area accurately reveals the original kitchen from **@Image1**.

## Scene 3 — 7–10 seconds

The final spray clears the remaining dirt from the floor and island. The worker lowers the pressure washer as the mist settles.

Reveal the kitchen completely clean, dry, illuminated, and identical to **@Image1**. Finish with a smooth cinematic pullback showing the entire spotless kitchen.

Ultra-realistic water physics, satisfying cleaning transformation, detailed grime removal, realistic reflections, no people besides the POV hands, no text, no subtitles, no logo, no watermark.

办公室角色讨论编程智能体

一个极短单行提示词,生成带有伪纪录片办公室情境的角色对白片段。

文生视频对白场景16:9请求时长: 10s
使用此提示词
查看完整原始提示词
Jim and Dwight from The Office discuss autonomous coding agents
精选

雨夜自行车一镜到底

雨夜街头的一镜到底,集中测试动作物理、道具连续性、霓虹文字和同步对白。

文生视频电影感16:9请求时长: 15s
使用此提示词
查看完整原始提示词
A 15-second photorealistic cinematic video captured as one continuous unbroken shot on a rain-soaked downtown street at blue hour.

A woman in a bright yellow raincoat rides a red bicycle quickly toward the camera while carrying a sealed paper coffee cup in one hand. [low-angle tracking shot]

She brakes sharply beside a glass storefront. The rear tire throws a realistic arc of rainwater across the pavement. The coffee cup slips from her hand, rotates once in the air, and she catches it cleanly without spilling. She then dismounts from the bicycle in one smooth, physically believable movement.

[smooth 180-degree camera orbit]

Maintain exactly the same woman, face, hairstyle, yellow raincoat, red bicycle, and coffee cup throughout the entire shot. Her body must move with realistic weight, balance, momentum, and contact with the bicycle and ground.

Behind her, a bright neon storefront sign remains clearly readable and correctly spelled throughout the shot:

“HAILUO 3.0 — ONE TAKE”

[slow cinematic push-in]

She looks toward the camera, smiles slightly, and says naturally:

“Okay... that was impressive.”

Photorealistic skin, believable hands, accurate reflections in the wet pavement and storefront glass, natural rain physics, realistic cloth and hair movement, cinematic lighting, shallow depth of field, and detailed environmental motion.

Native stereo audio: steady rainfall, bicycle chain movement, tire skid through water, coffee cup catch, distant city traffic, and perfectly synchronized dialogue.

No cuts, no scene transitions, no morphing, no identity changes, no wardrobe changes, no extra limbs, no disappearing objects, no misspelled text, and no background music.

恐龙多风格太空冒险

同一只恐龙连续穿越写实、科幻、二维动画、黏土和动漫风格,最终回到真人电影质感。

文生视频动画16:9请求时长: 15s
使用此提示词
查看完整原始提示词
A 15-second video that begins as a photorealistic cinematic dinosaur adventure and smoothly evolves through multiple visual styles without losing story continuity.

Scene begins in a misty prehistoric jungle at sunrise. A massive realistic Tyrannosaurus rex slowly steps into frame, its skin textured with scars, dust, and morning moisture. The camera tracks low beside its feet as the ground trembles with each step. Cinematic live-action style, natural lighting, realistic weight, realistic muscle movement.

[match cut transition]

The T. rex looks up. Its eye reflects the night sky, and the camera flies into the reflection, transitioning into outer space. The dinosaur is now floating beside a sleek silver spacecraft shaped like a fossilized bone. Earth is visible far behind it. Epic cinematic sci-fi style, dramatic lens flares, realistic stars, slow zero-gravity motion.

[liquid morph transition]

The scene transforms into a vibrant 2D animated style. The T. rex now wears a tiny astronaut helmet and comically pedals a rocket-powered bicycle across Saturn’s rings. Bright colors, expressive animation, smooth exaggerated motion, playful timing.

[paper rip transition]

The style changes into stop-motion clay animation. The dinosaur lands on a small asteroid, where tiny clay alien creatures offer it a glowing meteor sandwich. Tactile clay texture, handmade miniature sets, charming imperfect movement, visible stop-motion feel.

[glitch transition]

The style shifts into high-energy anime. The T. rex launches through a neon wormhole, roaring silently as planets streak past in colorful speed lines. Dynamic camera movement, intense anime lighting, dramatic perspective, cosmic scale.

[cinematic dissolve transition]

Final scene returns to photorealistic live-action cinema. The T. rex stands calmly on the moon in an astronaut helmet, looking at Earth. A small flag beside it clearly reads:

“HAILUO 3.0 TEST”

The camera slowly pushes in as the dinosaur blinks, exhales fog inside the helmet, and gently taps the moon dust with one claw.

Maintain the same dinosaur identity and recognizable silhouette across every style. Each transition should feel intentional, smooth, and visually creative. No random scene cuts. No disappearing dinosaur. No extra limbs. No unreadable text. No misspelled words. No chaotic camera shake. Make the final result feel like one connected mini-adventure through time, space, and animation styles.

竖屏家庭冲突微短剧

中国家庭冲突类竖屏短剧,以中近景、正反打和克制的真人表演推进情绪。

文生视频对白场景9:16请求时长: 15s
使用此提示词
查看完整原始提示词
A 9:16 vertical family-confrontation scene with grounded live-action performances, set in a Chinese family home or small restaurant. Use warm interior light, red decorations and calligraphy in the background, shallow depth of field, intense emotion, and tight pacing.

Performance: natural short-form drama, never theatrical. Qin Haoxuan argues back with anger, hurt, and urgency. The older woman questions him in a sharp, forceful, relentless tone. Build the confrontation steadily.

Shoot mainly in medium-close shots with frequent shot/reverse-shot cutting. Keep the setting lived-in and realistic. No sci-fi, period costume, or animation styling. Do not show subtitles, added text, platform watermarks, or stickers.
精选

MINIMAX 节拍同步动漫片头

六张图片和一段音频参考驱动 15 秒动漫片头,用精确时间段控制角色动作与动态图形转场。

参考生视频动画16:9请求时长: 15s
使用此提示词
查看完整原始提示词
Create a fast-paced 15-second, 16:9 anime opening-title sequence for an original series titled “MINIMAX.” Use Image1, Image2, Image3 as the overall visual, ensemble and color-style references. Use Image4 for the pink-haired swordswoman, Image5 for the red-haired swordswoman and Image6 for the black-haired swordswoman. Preserve each character’s face, hairstyle, outfit, proportions, colors and weapon design.

Use the supplied audio Audio1 as the exact timing reference. Synchronize every cut, character action, camera movement, graphic transition and title reveal to its beats, accents, fills, rises and final hit. Do not create a separate soundtrack. Add only subtle sword swishes, cloth movement and transition impacts under the reference audio.

VISUAL STYLE

A real high-energy anime opening combined with premium motion graphics. Use black, crimson, wine red, dusty pink, pale ivory and cool gray-blue. Combine angular split screens, diagonal masks, torn-paper shapes, manga-style framing, ink textures, petals, crimson moon forms, torii silhouettes, temple rooftops and sharp graphic lines.

Keep the sequence kinetic and animated, not a slideshow. Characters must run, turn, pivot, slide, draw their swords, change stance and react to momentum. Hair and clothing must remain in motion. Use tracking shots, whip pans, rapid push-ins, low angles and short camera orbits.

Never show multiple copies of the same character in one shot. A shot may feature one character alone or all three together naturally. No clone effects, repeated character fragments or layered duplicates.

SEQUENCE FLOW

[0.0–1.5s]
Open instantly on the first beat. A crimson line cuts through black as petals and white fragments burst outward. Pink, red and black graphic panels snap into place while a crimson moon quickly assembles in the background.

[1.5–4.3s]
Feature the pink-haired swordswoman alone. She runs into frame, plants one foot, turns and draws her katana in one continuous agile motion. Track beside her, then push toward her face. Cut between her eye, hand, blade and full-body movement on the audio accents. Use pink petals, red lines and diagonal panel cuts.

[4.3–7.2s]
Switch to the red-haired swordswoman on a strong musical hit. Use a low-angle tracking shot as she moves forward, pivots and performs one powerful controlled sword motion. Her long hair and garments sweep with momentum. Place a large crimson sun disc behind her while black-red panels strike into frame on the heavier beats.

[7.2–10.2s]
Feature the black-haired swordswoman alone. Begin with a moving close-up of her crescent earring and eyes, then orbit around her as she turns, changes stance and redirects her blade. Keep her hair and sleeves flowing. Use dark negative space, moon-shaped masks and precise cuts synchronized to smaller rhythmic details.

[10.2–12.7s]
At the main musical peak, reveal all three characters together in one shared scene. They enter through real movement and form a strong triangular composition. Use a fast camera push or short orbit as the crimson moon, torii silhouette, petals and graphic debris align behind them. Show each character only once.

[12.7–15.0s]
During the final phrase, alternate between very short moving close-ups of the three characters, one character per shot. Show an eye turn, hand gripping a sword, flowing hair or a blade catching light. On the final major hit, reveal the title “MINIMAX” through sharp diagonal masks, sword-line wipes and fractured graphic panels.

End on a clean ensemble composition with all three characters together. Keep subtle motion in their hair, clothing, petals and light reflections.

TRANSITIONS
Use beat-synchronized slash cuts, angular panel snaps, moon-mask wipes, petal streaks, torn-paper shutters, ink-impact cuts, foreground wipes and oversized title masks. No soft dissolves, slow fades, liquid morphing or random transitions.

TYPOGRAPHY
Use only the title “MINIMAX.” Keep it sharp, elegant, premium and fully readable. No credits, extra text, random symbols or broken lettering.

FINAL RESULT
A fast, polished and rhythmically precise anime opening with expressive character motion, strong graphic editing, strict visual consistency and no duplicated characters within the same shot.
精选

韩国公寓情感对话短片

韩国情侣对话短片,以身份锁定、屏幕方位、逐镜时间、表演调度和现场声音精细控制。

参考生视频对白场景16:9请求时长: 15s
使用此提示词
查看完整原始提示词
REFERENCE USE
Images 1 and 2 lock face, hair and wardrobe identity only. Do not copy their frontal pose, centred framing or lens gaze. Image 3 locks apartment geometry, light direction and materials only. Recompose the actors according to the screen directions below. Identities remain stable.

IDENTITY LOCKS
YEON-SU: Korean woman, 28. Loose unstyled dark hair tucked behind her anatomical right ear only; left ear covered. Oversized heather-grey knit, sleeves past both wrists, loose charcoal lounge trousers, bare feet, no jewellery.
JAE-HYUN: Korean man, 32. Dark knee-length coat stays on. Plain shirt open by exactly one button. Watch on anatomical left wrist. Keys handled only by anatomical right hand.
Never mirror these details. No wardrobe change.

SCENE
Three years into a relationship. The argument peaked before this clip. Neither is trying to win. YEON-SU needs him to deny he is leaving but cannot ask directly. JAE-HYUN has waited to be actively chosen and is ashamed that he needed it. His answer is an admission, not an attack. Her last line is not a reply; she discovers that she treated his staying as permanent while he experienced it as a choice she never made.

DIALOGUE
Spoken Korean only. Contemporary Seoul banmal between a long-term couple, with dropped subjects, intimate volume and natural consonants. Never translate, speak, subtitle or display English.

YEON-SU: "또 갈 거야?"
JAE-HYUN: "한 번쯤은... 가지 말라고 할 줄 알았어."
YEON-SU: "안 갈 줄 알았어."

SCREEN GEOGRAPHY
JAE-HYUN is frame-left facing screen-right. YEON-SU is frame-right facing screen-left. Singles leave look-room toward the unseen partner. Matched off-camera eyelines; never look into the lens. In the wide they are three metres apart. The table stays on YEON-SU's anatomical right so her right hand can hold it. The entry shelf stays beside JAE-HYUN's right hand. Fixed marks, but natural breathing, a swallow, finger pressure and tiny posture settling are allowed. No walking, sitting, approaching or contact.

SHOT LIST
[0.0-2.8] LOCKED WIDE TWO-SHOT, 35mm, eye level. Entry frame-left, living room frame-right. JAE-HYUN stands just inside the door, coat on. YEON-SU stands beside the table three metres away, right hand lightly holding its edge. She watches him; he looks at the shelf. At 1.3s he places the keys on wood with his right hand, quietly and deliberately. Neither speaks. Hold the distance.

[2.8-5.4] YEON-SU MEDIUM CLOSE, 75mm, three-quarter view, mid-chest to head. She is frame-right with look-room screen-left. Begin neutral, without prepared sadness. One ordinary shallow breath. Between 3.7s and 4.6s she asks, "또 갈 거야?" Low, plain, almost practical. She wants him to say no. Brows stay relaxed; mouth closes after the line. She listens.

[5.4-10.3] JAE-HYUN MEDIUM CLOSE, 75mm, three-quarter view, mid-chest to head. He is frame-left with look-room screen-right. Mouth fully closed during the silence. Eyes down; one controlled breath and one swallow. This is active decision, not blank posing. He raises his eyes only when ready to admit it. Between 7.9s and 9.6s: "한 번쯤은... 가지 말라고 할 줄 알았어." Quiet, warm, tired. The pause after "한 번쯤은" is a failed attempt to make the sentence easier. No blame, emphasis or sigh. He keeps his eyeline toward her afterward.

[10.3-15.0] YEON-SU CLOSE, 85mm, collarbones to crown, three-quarter view, frame-right with look-room screen-left. One continuous take with no reframing or hidden cut. She is already looking down. Begin without a hurt expression. Recognition forms through stopped breath, a released jaw and eyes focusing on nothing. No tears. Between 12.7s and 13.9s she says to the floor, "안 갈 줄 알았어." Nearly matter-of-fact at first; the final word softens when she hears her own assumption. Only after the last syllable does she lift her eyes toward him. Mouth closed. Hold her listening face through 15.0s.

ACTING
Lived-in screen acting, never presented emotion. Each actor actively listens to the absent partner. Emotion appears through attention, breath and thought, not facial display. Allow imperfect human timing inside the dialogue windows. No added theatrical pauses or mirrored performances.

LIGHT AND IMAGE
One low warm living-room practical, about 2900K. Weak cold night spill from the dark entry, not saturated cyan. Faces partly in shadow; natural Korean skin tone, pores and under-eye texture. Desaturated wood, grey knit and blue-black exterior; restrained grain. The lamp motivates the warm face light. No commercial orange-teal grade, bloom, halation, glow or flare.

CAMERA
Locked and eye-level. Maintain the 180-degree line. Singles are three-quarter views, never centred frontal portraits. Shallow depth of field with both eyes readable. No drift, push-in, handheld sway, arc, rack focus, dutch angle, motion blur or lens-scale change within a shot.

PRODUCTION SOUND
Natural Korean production dialogue recorded in the apartment, not narration, dubbing or synthetic TTS. Intimate conversational level, slight room reflection, matched boom-and-lav perspective. Continuous stereo room tone beneath every line and silence: refrigerator hum, low HVAC and distant traffic through closed glass. Never hard-gate the room. Keys on wood at 1.3s are the only featured effect. No phone buzz or designed silence drop.
No music, score, drone or tonal swell. Dialogue centred; ambience has subtle width. No breath boosting, over-compression or loud social-video mastering. If controllable: 48 kHz stereo, true peak at or below -6 dBFS.

NEGATIVES
No English voice, subtitles, captions, text, logos or watermarks. No lens gaze, centred portrait composition, silent mouthing, speech outside the three lines or bad lip sync. No shouting, sarcasm, crying, tears, sobbing, gasping, widened eyes, knitted brows, trembling lips, clenched fists, pointing, arm crossing or hand gestures. No extra people or reflections, wardrobe drift, side mirroring, beauty filter, face warp or identity change. No extra cuts, jump cuts, punch-ins or framing resets, especially from 10.3s to 15.0s.

深夜高速灯轨延时摄影

仅用两行日文提示词生成深夜城市高速延时摄影和持续的车灯残影。

文生视频电影感not-stated请求时长: 未注明
使用此提示词
查看完整原始提示词
・深夜都市高速のタイムラプス
・車のライトの残像が残るタイムラプス

繁茂花园里的死神

用四个英文单词测试 H3 对强烈主体与环境反差的表现。

文生视频奇幻not-stated请求时长: 未注明
使用此提示词
查看完整原始提示词
grim reaper in lush garden
精选

彩色喷漆武器选择游戏

参考图驱动的游戏菜单依次切换喷漆武器、确认装备,并无缝进入第三人称玩法。

参考生视频游戏与 UInot-stated请求时长: 15s
使用此提示词
查看完整原始提示词
Use Image 1 for the character, Image 2 for the UI style, and use the uploaded weapon selection image as the reference for all paint weapons. The selectable weapons should closely match the colorful ink guns shown in the reference image, including compact paint shooters, dual paint pistols, heavy paint cannons, bucket/slosher weapons, large paint brushes, paint rollers, sniper-style paint rifles, and other stylized colorful ink weapons. Match their playful proportions, vibrant colors, chunky plastic construction, and game-like appearance while avoiding exact logos or branding.

[0–2 seconds]
High-angle overhead shot. The character stands on a vivid, highly saturated purple floor, looking up toward the camera, matching Image 1 exactly. A colorful game menu matching Image 2 appears on the right: START NEW GAME, CONTINUE (highlighted), SETTINGS, EXIT GAME. Player profile MINIMAX appears in the top left. The cursor selects CONTINUE.

[2–4 seconds]
The camera pushes toward her right hand. A PAINT WEAPON SELECT panel slides in, using the uploaded weapon selection screen as its layout and visual reference. Multiple colorful paint weapons are displayed exactly like the reference image in a grid with rarity and level indicators. The cursor moves between different weapons. Every time a weapon is highlighted, the paint gun in her hands instantly transforms into the selected model with colorful ink particles, playful mechanical animation, spinning parts, and glossy paint effects. She remains standing throughout.

[4–7 seconds]
The camera smoothly arcs around her left side. The selector rapidly cycles through many weapons from the reference image: compact paint shooters, oversized paint brushes, dual paint pistols, heavy paint cannons, bucket-style splash weapons, sniper paint rifles, giant paint rollers, and experimental ink launchers. With each selection, the weapon in her hands changes immediately to match the highlighted weapon while she naturally adjusts her grip. Every swap is accompanied by bursts of colorful ink, animated UI effects, splashing paint particles, and satisfying transition animations.

[7–8.5 seconds]
Pull back to a medium shot. CONFIRM LOADOUT flashes. The cursor clicks it. All UI panels collapse and disappear. She confidently spins the selected paint weapon once before resting it on her shoulder, remaining in a relaxed standing pose.

[8.5–10 seconds]
A colorful LOADING bar fills from 0% to 100%. The purple testing room dissolves beneath spreading ink splashes that transition into the game world.

[10–15 seconds]
The world loads into a bright, colorful ink-covered city filled with graffiti, paint-coated streets, animated billboards, moving NPCs, and playful architecture. The camera settles into a third-person view behind the character. HUD elements fade in: minimap, health, paint gauge, current weapon icon matching the selected weapon, ammo counter, and objective marker. She runs forward while firing the selected colorful paint weapon, coating walls, streets, obstacles, and scenery with thick, glossy paint that matches the weapon's color.
精选

指甲上的微缩生命世界

五枚指甲分别化作锦鲤池、樱花、星空、森林和微型龙的动态入口。

文生视频奇幻not-stated请求时长: 未注明
使用此提示词
查看完整原始提示词
A stunning, perfectly manicured female hand rests elegantly on a smooth, black marble surface, her fingers slightly spread to showcase each intricately designed, living fingernail. The camera remains perfectly still, ensuring an uninterrupted, high resolution full frame view of the hand and its enchanting animated nail art. The soft, warm ambient lighting casts subtle shadows across the textured marble, enhancing the hyper-realistic depth of her flawless, porcelain smooth skin.

Each fingernail is a miniature portal of magic, alive with independent yet synchronised micro scenes:

Thumb Nail  A tiny koi pond, where goldfish gracefully swim beneath a translucent, glass like surface, their scales shimmering as they dart between delicate lily pads swaying gently in the water’s micro currents.

Index Finger: A blooming cherry blossom tree, where soft pink petals flutter in an invisible breeze, some occasionally detaching and dissolving into golden light.

Middle Finger: A miniature celestial night sky, where tiny constellations shift and twinkle, meteors streaking across the deep, cosmic expanse, casting faint, dancing reflections on the nail’s glossy surface.

Ring Finger: A tiny, enchanted forest, where glowing fireflies hover and flicker, their gentle luminescence casting soft pulses of golden light across the leaves, which sway subtly as if responding to the heartbeat of the world within.

Pinky Finger: A coiled, ethereal dragon, its microscopic, iridescent scales shifting in colour from deep crimson to molten gold. It blinks once, then exhales a swirling wisp of blue fire, which vanishes the moment it drifts beyond the nail’s edge.

Her fingers remain completely still, yet the nail animations play simultaneously, creating a mesmerising symphony of motion within a perfectly composed frame. The lighting subtly shifts, casting gentle reflections off the glossy nails, reinforcing their three dimensional realism.
精选

兽人与游侠的酒馆格斗

15 秒参考角色酒馆格斗,重点约束惯性、碰撞、恢复动作和环境物理反馈。

图生视频奇幻16:9请求时长: 15s
使用此提示词
查看完整原始提示词
Style: Hyper-realistic dark fantasy tavern cinematic, grounded physical combat, realistic body momentum, medieval atmosphere, gritty lighting, practical effects, handheld cinematic camera, physically accurate movement, ultra-detailed textures, realistic collisions and environmental interaction.

Duration: 15 seconds

Aspect Ratio: 16:9

IMPORTANT:

Keep the SAME characters fully consistent with the reference: 📷Image1

— Nyssa: athletic woman, dark curly hair tied back, bronze leather armor, cloth wraps, sword on hip

— Grok: massive muscular green orc, scarred skin, fur armor, heavy axes, large physical weight

Combat MUST obey realistic physics:

— no floating

— no anime flips

— realistic inertia and recovery

— believable weight transfer

— realistic impacts and exhaustion

— grounded footwork

— environment reacts physically to impacts

[00:00-00:02]

Wide cinematic shot inside a crowded medieval tavern filled with drunk mercenaries, candles, smoke, spilled ale, and loud cheering. Nyssa walks through the tavern cautiously while patrons stare. Grok sits at a massive wooden table drinking heavily. Deep bass-heavy tavern music and crowd ambience.

[00:02-00:04]

Close-up tension sequence:

— Nyssa locks eyes with Grok

— Grok slowly stands, towering over everyone

— benches scrape across the floor

— mugs shake from his weight

— crowd forms a fighting circle cheering loudly

Handheld camera movement feels realistic and grounded.

[00:04-00:07]

The fight erupts violently. Grok swings a heavy punch with believable momentum. Nyssa narrowly dodges while stumbling realistically into a wooden table. Wood cracks and splinters physically on impact. She counters with fast grounded strikes to Grok’s ribs and legs. Crowd roars and throws coins and mugs into the air.

[00:07-00:10]

Combat intensifies through the tavern:

— Grok grabs Nyssa and throws her across a table

— table explodes realistically beneath her weight

— Nyssa rolls across the floor recovering naturally

— Grok charges heavily, smashing furniture

— patrons jump away realistically

No exaggerated movement, only physically believable combat.

[00:10-00:12]

Nyssa uses speed and positioning intelligently. She dodges another heavy attack causing Grok to crash into a support beam. Dust and debris fall from the ceiling. She climbs briefly onto the bar counter and leaps down with controlled realistic momentum, locking Grok’s arm and using leverage to throw him off balance.

[00:12-00:15]

Final cinematic climax. Nyssa tackles Grok to the ground hard onto broken wooden debris. The floor shakes from the impact. She pins him down on top of him while holding a blade near his throat. Grok struggles realistically beneath her massive weight difference and exhaustion. Tavern crowd erupts cheering wildly, slamming mugs against tables. Camera slowly pushes inward on Nyssa breathing heavily while sweat, dirt, and candlelight flicker across both fighters.

Audio:

Heavy cinematic tavern battle soundtrack with deep drums, loud cheering crowds, breaking wood, metal impacts, realistic body hits, roaring fire ambience, mugs crashing, heavy footsteps, and gritty medieval atmosphere.

Negative prompts:

No floating, no anime combat, no superhero physics, no unrealistic flips, no weightless motion, no blurry characters, no low detail, no cartoon style, no slow reactions, no clipping through objects.

诡异霓虹汉堡广告

以 90 年代惊悚片美学结合快节奏美食特写和诡异喜剧,制作汉堡广告。

文生视频商业广告not-stated请求时长: 15s
使用此提示词
查看完整原始提示词
Cinematic 15-second commercial shot, fast-paced editing, dark humor, creepy comedy style. Ultra-realistic, 8k resolution. A surreal hamburger restaurant with moody neon lighting (red and green). Close-ups of distorted, wide-eyed hungry people smiling creepily while eating massive juicy cheeseburgers. Fast cuts: sizzling bacon in flames, towering stacks of burger trays, a mysterious retro waiter with a unnerving wide smile. Hyper-detailed melted cheese, glossy sauce drips, chaotic and fun atmosphere, 90s thriller aesthetic crossed with a fast-food ad.

霓虹雨中的赛博武士

连续慢动作环绕镜头拍摄赛博武士在强烈霓虹雨中缓慢拔出发光武士刀。

文生视频电影感not-stated请求时长: 15s
使用此提示词
查看完整原始提示词
Continuous 15-second cinematic shot, extreme slow-motion: a futuristic cyberpunk samurai standing in heavy neon rain, reflective wet armor. The samurai slowly draws a glowing katana, lens flare, volumetric neon fog, dynamic camera panning around the character, continuous fluid motion, photorealistic 8k

2000 年代动画风上学问候

短英文提示词要求 2000 年代动画风少女在上学路上看向观众并用日语说“早上好”。

文生视频动画not-stated请求时长: 未注明
使用此提示词
查看完整原始提示词
This is an anime scene in the style of the 2000s. A girl is walking along her usual route to school. As she passes by, she looks toward the viewer and says, "おはよう"

芝加哥街头服装多地点快切

商业创意方案使用六个芝加哥地点、快速切换和最终主角推进,同时锁定服装与人物气质。

参考生视频商业广告not-stated请求时长: 未注明
使用此提示词
查看完整原始提示词
Creative Direction:
Use a high-impact opening, six fast location changes, and a final hero push-in to turn a simple streetwear walk into a complete visual journey through Chicago.

Open with a bold establishing shot, then move through a rapid sequence of distinct Chicago locations while keeping the character, wardrobe, and attitude completely consistent.

Art Direction:
Downtown streets, elevated trains, storefronts, concrete architecture, hard cuts, dynamic tracking shots, modern streetwear, natural city light, and a polished urban-editorial finish.

Train platforms, wide-angle movement, whip-fast transitions, layered street styling, realistic motion, and crisp cinematic pacing.

伪纪录片办公室政治独白

15 秒伪纪录片对白提示词,精确安排三名角色反应、办公室环境声和冷幽默结尾。

文生视频对白场景not-stated请求时长: 15s
使用此提示词
查看完整原始提示词
The Office — Michael admires Trump. 0:00–0:04 — Medium shot, Dunder Mifflin bullpen, documentary handheld. Michael stands by Jim and Pam’s desks, hands animated, big proud smile. Jim pretends to work. Pam holds a mug. Soft office chatter, phones, AC hum. No music. 0:04–0:09 — Michael turns toward camera slightly and says brightly: “I’m just saying — Trump? Total winner. Very stable genius. Like me, but with better hair.” 0:09–0:12 — Jim slowly looks up from his monitor, deadpan, eyebrows raised. Pam freezes mid-sip, stares at Michael with a tight, confused smile. Awkward silence. A phone rings once in the background. 0:12–0:15 — Michael keeps nodding, totally oblivious, double thumbs-up. Jim and Pam exchange one weird glance. Hard cut on their shared “what did he just say” faces. Style: photoreal The Office mockumentary, subtle handheld shake, short clear dialogue, office room tone only, no music, no subtitles, no on-screen text, no logos.

提示词数据集:Awesome Hailuo H3 Real-World Video Prompts,经改编与转存。各条提示词、视频、角色与商标权利仍归原创作者所有。许可协议 CC BY 4.0

MiniMax H3 提示词公式

MiniMax H3 提示词公式:先写清画面中发生什么,再安排镜头与声音

好用的 MiniMax H3 / 海螺3提示词,本质上是一份精简的拍摄任务书。只写本镜头真正需要的部分;时长、比例、分辨率等生成参数应在界面中设置。

主体 + 动作或事件 + 场景 + 视觉表达 + 运镜或切镜 + 声音

主体

说明人物、商品、动物或物体是谁;只补充必须保持稳定的身份特征。

动作或事件

写清画面发生的变化。每个阶段只安排一个主要事件,并写出可见的结束状态。

场景

交代地点、时间、天气、背景状态与重要的空间关系。

视觉表达

指定光线、色彩、材质、质感、写实程度与整体情绪。

运镜或切镜

选择景别、机位、运动方向、对焦对象、转场或镜头顺序。

声音

只有声音影响成片时,才写对白、音色、环境声、动作音效或音乐。

完整 MiniMax H3 文生视频提示词

一名陶艺师在清晨安静的工作室里完成一只浅蓝色陶杯,将它从转盘上取下并放到木架中央。窗外柔和晨光照入室内,湿润陶土呈现细腻光泽。镜头先以中景记录双手拉坯,再缓慢推近杯口纹理,最后切到成品正面。保留转盘低沉转动声、陶土摩擦声和轻微室内环境声。

海螺3提示词写法

MiniMax H3 提示词怎么写:让 H3 视频更可控的 4 个关键技巧

这些海螺3提示词写法把创意拆成可以逐镜头检查的指令,无论你走文生视频、图生视频还是参考生视频,修改时都更容易定位问题。

1

逐份说明参考素材职责

明确哪张图负责五官、服装、商品、场景或材质;如果背景或人物容易被误带入,也要说明不采用什么。

@图片1定义<主厨>的五官、短发和深蓝围裙,不采用图片背景。@图片2定义餐厅后厨的空间布局与暖色顶光,不采用图中人物。让<主厨>在出餐台完成一道菜的摆盘。
2

长视频按阶段组织

每个阶段只发生一次主要变化,结尾写成可以直接看到的状态,并把它带入下一阶段。

阶段一:<花艺师>修剪花枝;结束时花束握在她左手,剪刀放在桌面右侧。阶段二:<店员>包装同一束花并系上绿色丝带;结束时花束平放在桌面中央。阶段三:<店员>把花束放到取货架。
3

关键节拍再用时间戳

时间戳适合进出场、交接、转场和明确节拍。区间要连续,不要要求一秒完成多个复杂动作。

0–4秒:空置展示台;一只手把白色陶盘放到中央后离开。4–9秒:移走陶盘,在同一位置放下透明玻璃杯。9–15秒:玻璃杯被绿色陶瓶替换,最后缓慢推近陶瓶。
4

把摄影术语翻译成可见变化

专业运镜同时写明作用主体、起点、方向、速度和最终景别,比只丢一个术语更稳定。

对坐着的宇航员使用希区柯克变焦:保持她的面部大小不变,同时让后方走廊向远处拉伸。焦点从有划痕的头盔面罩平滑转移到双眼,最后停在稳定特写。

可复用 MiniMax H3 与海螺3提示词模板

免费海螺3提示词模板:首尾帧、分镜、视频续写与原生音频

复制一份 MiniMax H3 提示词模板,替换尖括号中的内容,只保留当前生成模式真正需要的段落。

图生视频 / 首尾帧

锁定开场与落点状态

@图片1作为首帧,定义<开始构图、主体姿态、道具状态和镜头方向>。@图片2作为尾帧,定义<最终构图和可见结束状态>。画面通过<主要动作>从图片1连续过渡到图片2。全过程保持<主体身份、服装、道具结构、场景布局、光线和运动方向>一致。
宫格分镜 / 多镜头视频

把分镜图改写成镜头清单

@图片1提供<N格分镜>的镜头顺序,按照从左到右、从上到下读取;不采用线稿画风与文字标注。镜头1:<景别与动作>。镜头2:<景别、动作和转场>。镜头3:<细节特写>。最终镜头:<收束动作与可见结束状态>。最终画面采用<视觉风格>,声音包括<对白、环境声、音效或音乐>。
视频向后续写

先承接边界画面,再发生新事件

@视频1是需要向后续写的原视频。生成片段的第一个画面直接承接它的尾帧:保持<姿态与朝向>、<道具位置>、<背景>、<构图>、<光线>、<声音状态>和<运动趋势>连续。随后发生<新动作或事件>。同一主体始终是一个连续对象,不重复、不分裂、不换身份。
对白与原生音频

把台词绑定到说话人与表演

一名年轻女子在车站入口停下脚步,回头,用自然的洛杉矶美式英语和克制但如释重负的语气说:{You actually made it.} 保留远处列车刹车声、轻微脚步声与柔和的车站环境声;不要字幕,不要画外旁白。

海螺3提示词排错

MiniMax H3 与海螺3提示词常见问题及快速修复

人物或道具在镜头间互换

给每个主体命名并绑定参考图,写清道具归属,再在下一阶段开头重复上一阶段的结束状态。

自动切镜太多

每个阶段只保留一个主要事件,把形容词堆砌改成可见动作;需要连续镜头时直接写“一镜到底”。

参考图带入错误背景

同时写正向职责与排除项,例如“只采用五官和服装,不采用房间、姿势与机位”。

情绪表演太泛

把情绪词换成 2–4 个可见信号,例如视线、呼吸、肩膀、嘴角、手部与说话节奏。

MiniMax H3 与海螺3提示词常见问题

MiniMax H3 提示词的最佳公式是什么?

从主体和动作开始,再按需补充场景、视觉表达、运镜或切镜与声音。重要的身份和连续性约束要靠前;不影响画面的部分可以省略。

海螺3提示词和 MiniMax H3 提示词有区别吗?

“海螺3提示词”是创作者寻找 MiniMax H3 视频提示词时常见的搜索说法。本页中两者指向同一套 H3 实用写法,并不代表两套不同的提示词语法。

MiniMax H3 图生视频提示词怎么写?

先说明原图负责哪些人物、构图或材质,再写新增动作和运镜,并明确哪些内容必须保持。如果提供尾帧,还要描述最终应该落到的可见状态。

需要写负面提示词吗?

优先使用正向、可观察的描述。排除项主要用于防止参考素材误带背景、额外人物、身份互换、自动字幕,或视频编辑时误改目标之外的内容。

什么时候应该使用时间戳?

关键节拍、人物进出场、交接与转场适合时间戳。普通叙事优先使用“阶段 + 结束状态”,更容易撰写和修改。

这些 MiniMax H3 提示词模板可以修改吗?

可以。替换主体、动作、场景、镜头和声音,同时保留模板结构即可。示例是起点,不是必须逐字照抄的命令。

这些 MiniMax H3 提示词免费吗?

本页所有提示词都可以免费阅读、复制和改编。它们来自创作者的公开帖子,因此转发成片时请署名创作者,并尊重提示词中涉及的品牌或第三方角色权利。

MiniMax H3 能直接用提示词生成带声音的视频吗?

可以。MiniMax H3 在一次生成中同时产出画面与音频,对白、环境声、音效和音乐都来自提示词本身。把每句台词绑定到指定说话人,并写清语气、口音和环境声,而不是后期再加。

海螺3视频最长多少秒?分辨率是多少?

MiniMax H3 单段视频为 4 到 15 秒,分辨率为 768P 或 2K。两者都是界面设置,所以时长、画幅和分辨率不要写进提示词。需要更长内容时分段生成,再用参考素材保持连贯。

MiniMax H3 提示词最多能带几个参考文件?

最多 9 张图片、3 段视频和 3 段音频,合计 12 个。数量不是越多越好:两张长相不一致的人脸参考会让模型在两者之间摇摆。每项职责用一份干净参考,并写明每个文件负责什么。

把一条 MiniMax H3 提示词变成你的下一个镜头

先写一个可见事件,只补充真正影响结果的镜头与声音,然后根据生成结果继续迭代。本页每条海螺3提示词都可以免费复制改编。