Hailuo AI Prompt Generator: Ready-to-Paste Prompts for MiniMax H3

Hailuo has always been the model people pick for physics: water that splashes like water, fabric with weight, bodies that land hard. MiniMax H3 (often called Hailuo 3.0, though MiniMax’s official name is H3) kept that and added native stereo sound, 2K output and full multimodal references. It also kept Hailuo’s odd little superpower, square-bracket camera commands. This Hailuo AI prompt generator page shows how to write for H3 specifically: a formula, nine ready prompts, and the reference wording MiniMax’s own docs use.

What MiniMax H3 Is Good At Right Now (and Its Limits)

From MiniMax’s launch post and API documentation:

• Released: July 31, 2026, as a single model that reads text, images, video and audio together.
• Length: 5 to 15 seconds in the Hailuo app (the API accepts 4 to 15, whole seconds only).
• Resolution and frame rate: up to 2K (1440p) at 24 fps, with 768P as the cheaper option. The H3 Max variant (built with fal) runs at 480P or 768P.
• Aspect ratios: 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16.
• Audio: native stereo generated with the clip: dialogue, effects, room tone. It can also follow a voice from a reference recording.
• References: up to 12 files per generation: 9 images, 3 videos and 3 audio tracks.
• Price: MiniMax says H3’s per-second price at 2K is less than a third of mainstream models.
• Open weights: MiniMax announced it would release the model weights, which makes H3 unusual among top video models.

The limits: 15 seconds is the ceiling per generation, and there is no multi-shot label system like Kling’s. For labeled cuts, the Kling prompt generator fits better. For 30-second takes, use Seedance.

The MiniMax H3 Prompt Formula

H3 reads natural language, with camera commands dropped in square brackets right after the moment they apply to.

[Scene: place, time, light]. [Subject] [action with physical detail] [Camera command]. [Second action] [Camera command]. [Look: style, lens, grade]. Sound: [dialogue, effects, ambience].

With references, name each file by its index and give it a role:

Use Image 1 as the character. Use Video 1 as the camera reference and follow its path.
[What the character does, in physical detail]. [Look]. Sound: [what we hear].

The camera command set carried over from earlier Hailuo models: Push in, Pull out, Zoom in, Zoom out, Pan left/right, Truck left/right, Tilt up/down, Pedestal up/down, Arc shot, Tracking shot, Static shot, Shake. Combine moves inside one bracket: [Truck left, Pedestal up].

9 Copy-Ready MiniMax H3 Prompts

CCTV: dog vs pool flamingo

Static CCTV camera overlooking a backyard swimming pool on a sunny afternoon, slightly faded colors, mild fisheye distortion [Static shot]. A golden retriever sprints across the deck, leaps onto a giant inflatable flamingo, and the flamingo flips over, launching the dog into the water with a huge splash. The dog surfaces and paddles in circles, delighted. Security camera aesthetic, compression artifacts. Sound: paws on wet tile, a big splash, happy panting.

Why it works: the flip and splash give H3’s physics something real to simulate, and [Static shot] keeps it looking like genuine surveillance footage. More setups in CCTV prompts.

Pet camera: drawer burglar

Indoor pet camera footage, wide angle, slightly blurry, evening lamp light in a small kitchen [Static shot]. A black cat paws a kitchen drawer open inch by inch, climbs inside, and pulls the drawer almost shut behind itself with one paw, leaving only its tail sticking out. Grainy home camera look. Sound: drawer rail sliding, a soft thump, a faint chirp from inside the drawer.

Why it works: the “inch by inch” detail slows the action down, and the tail left sticking out gives the viewer a payoff shot. Find more in pet video prompts.

Handheld phone fail: the kayak launch

Handheld phone video from a wooden lake dock on a bright summer morning, slightly shaky [Shake]. A man in sunglasses steps confidently from the dock into a yellow kayak, it rolls sideways, and he tips into the lake in one smooth motion, sunglasses still on. Water sprays across the lens. His friend filming laughs. Natural sunlight, phone exposure adjusting. Sound: wood creaking, a heavy splash, laughter. Friend: "Ten out of ten entry."

Why it works: the roll-and-tip is exactly the kind of weight transfer Hailuo handles well, and the lens spray sells that it was filmed on a phone.

Oddly satisfying: kinetic sand slice

Extreme close-up on a smooth block of pastel purple kinetic sand on a white table, soft even overhead light [Static shot]. A thin metal knife presses slowly down through the block and the slice peels away and crumbles in a soft cascade [Push in]. Macro lens, shallow depth of field, crisp detail. Sound: soft crunch, fine grains trickling.

Why it works: H3’s 2K output has room for grain-level detail, and a slow push-in during the crumble holds on the satisfying moment. See oddly satisfying prompts.

Cinematic: fishing boat in a storm

A small wooden fishing boat climbing a huge dark wave at dusk, rain slanting through the frame [Truck left, Pedestal up]. An old fisherman in orange oilskins grips the wheel as spray crashes over the bow and drains across the deck. Lightning briefly lights the sea from behind. Desaturated blue-green grade, shot on 35mm, heavy grain. Sound: roaring wind, waves slamming the hull, creaking wood.

Why it works: combining two moves in one bracket gives a rising, sideways reveal of the wave without a cut.

Product ad: soda can splash

A cold red soda can on a black reflective surface, condensation beading on the metal, a single hard backlight [Static shot]. Ice cubes drop around the can and splash cold water outward in slow motion, droplets freezing mid-air [Arc shot]. Glossy commercial look, high contrast, crisp 2K detail. Sound: ice clinking, a can crack and fizz at the end.

Why it works: starting static and moving into an arc on the splash puts the camera move exactly on the money shot.

Security camera horror: hospital hallway

Black-and-white security camera footage of an empty hospital hallway at 3 AM, flickering fluorescent tubes, high angle from the ceiling corner [Static shot]. At the far end, a wheelchair slowly rolls out of a doorway by itself, stops in the middle of the hallway, and turns to face the camera. The lights flicker once. Heavy grain, rolling interference lines. Sound: electrical buzz, a slow squeaking wheel, then silence.

Why it works: a slow, physical event plus a sound drop-out does the scaring, no monster required. More in horror prompts.

Reference: character plus camera path

Use Image 1 as the character. Use Video 1 as the camera reference and follow its exact camera path. The character, a small robot with a cracked blue visor, walks through a rainy neon street market, stops at a noodle stall, and tilts its head at the steam rising from a pot. Photoreal, wet reflections, 85mm lens feel. Sound: rain on awnings, sizzling woks, soft servo whirs.

Why it works: this follows MiniMax’s own guidance: when a video sets the camera path, describe the subject and leave the move out of the text so the two do not conflict.

Reference: character sings to your audio

Use Image 1 as the character: a cartoon hamster in a tiny leather jacket. Use Audio 1 for the vocals. The hamster stands on a tin-can stage under a desk lamp spotlight, grips a pencil like a microphone, and sings with big dramatic head tilts, lip-synced to Audio 1 [Push in]. Warm, cozy 3D animation style. Sound: the vocals from Audio 1, a small crowd of squeaks cheering at the end.

Why it works: it mirrors MiniMax’s launch example (a character from an image singing to a reference track), and naming both files gives each a single role.

MiniMax H3-Specific Tips

• Put brackets right after the action they belong to. That is where MiniMax’s docs place them, and it ties each move to a specific moment.
• Do not double-direct the camera. With a camera reference video, remove movement words from the text.
• Watch reference audio. A reference video carries its sound too, so the output voice tends to follow it. State the voice you want in the Sound: line if it should differ.
• Draft at 768P, finish at 2K. Get the action right cheaply, then rerun the keeper at full resolution.
• Write physics as a sequence. “Rolls sideways, tips in, water sprays the lens” beats “falls in the water.” The existing MiniMax Hailuo guide has more on physical detail.

From One Idea to 10 Hailuo Prompts With LzyPrompt

The slow part of making Hailuo content is not the brackets. It is coming up with ten good versions of one idea. In LzyPrompt you type something like “dog flips a pool float,” choose a Camera Style (CCTV security camera, handheld phone video, doorbell camera and others) or keep Auto-detect, and set Shot Type, Lighting or Aesthetic if you care. You get 10 variations, each with a title, a full prompt with setting, lighting, action beats, imperfections and sound cues, and a ready caption for Shorts or Reels.

LzyPrompt works with any model and has no Hailuo mode, so the H3 finishing touches are yours: add a bracket command after the key action, rename the sound cues into a Sound: line, and add Image 1 or Video 1 roles if you are using references. When you want a week of uploads in one go, export all 10 to CSV. You can try LzyPrompt free for 7 days.