Hailuo Prompting Guide: Prompts for Natural AI Video
A good Hailuo prompt is a short scene description in this order: subject, one clear action, camera movement, setting and light, then a pace word such as "slowly". Keep it to two or three sentences, use film vocabulary for the camera, and leave out quality tags like "8k" that the model does not act on.
Hailuo (hailuoai.video, by MiniMax) has a reputation for believable human motion and smooth camera work. Those strengths show up only when the prompt gives the model something to act on. Vague prompts get vague, slightly floaty clips; prompts written like a shot description from a script get clips that look directed. This guide covers the structure that works for text-to-video and image-to-video, the vocabulary Hailuo responds to, and the habits that waste credits.
How Hailuo reads a prompt
Hailuo generates clips of a few seconds from text alone or from a first-frame image plus text. Several model versions exist in the current interface, and the newer ones follow prompts more literally, which means your wording matters more than it did a year ago. Some versions offer an optional prompt enhancement that rewrites your text before generation; it helps beginners but removes control, so turn it off once you know what you want.
The model is good at a few things worth leaning on: natural body movement, fabric and hair, weather, water, and gradual camera moves. It is weaker at text, precise hand actions, more than two characters interacting, and anything that needs a fast cut. Write prompts that play to the first list.
The prompt structure
Use this order. It is the order a cinematographer would describe a shot, and it maps cleanly onto what the model needs.
- Subject. Who or what, with two or three identifying details. "A grey-haired fisherman in a dark wool jumper."
- Action. One verb phrase. "Coils a rope on the deck." Not a sequence of actions.
- Camera. "Slow dolly in", "static wide shot", "handheld, following from behind". Say it even when you want the camera still; otherwise the model chooses for you.
- Setting and light. "Small harbour at dawn, soft overcast light, wet stone." Light sets mood more than any adjective does.
- Pace and texture. "Slow, calm movement, natural motion." Pace words are the single most useful addition to a Hailuo prompt.
Assembled: "A grey-haired fisherman in a dark wool jumper coils a rope on the deck of a small boat. Slow dolly in. Small harbour at dawn, soft overcast light, wet stone. Slow, calm movement." Three short sentences, nothing wasted.
For image-to-video, drop the subject, setting and light (they are in the image) and keep only action, camera and pace. The Hailuo image to video guide covers that workflow.
Vocabulary that works
- Camera moves: static shot, slow push in, pull back, pan left or right, tilt up or down, tracking shot, orbit, handheld drift, crane up. Use one per clip; two if the second is subtle.
- Shot sizes: wide shot, medium shot, close-up, over-the-shoulder. These change framing dramatically on text-to-video.
- Pace: slowly, gently, gradually, subtle, calm, drifting. Fast words ("rushes", "whips") tend to produce artefacts.
- Light: golden hour, overcast, backlit, window light, neon, firelight, fog. Name the source, not just the mood.
- Motion detail: hair moves in the breeze, fabric sways, steam rises, ripples spread. Small secondary motion makes clips feel alive without destabilising the subject.
Words to drop: "8k", "ultra-detailed", "masterpiece", "trending". They are image-model habits and do nothing for Hailuo. Brand names, living artists and real people are also out, both for policy reasons and because they produce unpredictable results.
Settings at a glance
| Setting | What it does | Start with |
|---|---|---|
| Model version | Which Hailuo model generates; newer versions follow prompts more literally | The latest version offered |
| Mode | Text-to-video or image-to-video (first frame) | Image-to-video when composition matters |
| Prompt enhancement | Platform rewrites your prompt before generation, where offered | On for a first try, off for control |
| Duration | Clip length | Shortest option until the motion is right |
| Resolution | Output size where the version offers a choice | Standard for drafts, highest for the keeper |
| Subject reference | Uploaded image that fixes a character’s appearance across clips | One clear, front-facing reference |
Labels and options vary by model version and change between releases; the current interface is the authority.
Prompt patterns for common jobs
- Product beauty shot: "A matte black ceramic bottle on a stone plinth. Slow orbit around the bottle. Soft studio light from the left, dark background. Steady, slow movement." Keep the label away from the camera path; lettering will not survive a full rotation.
- Portrait with life: "A young man in a denim jacket looks down, then slowly raises his eyes to camera and half-smiles. Static medium close-up. Window light, warm interior. Subtle natural motion."
- Atmosphere plate: "Mist drifts across a pine forest at dawn. Slow crane up through the trees. Cool blue light, soft shadows. Gradual, calm movement." Plates like this cut well under titles.
- Character walk: "A woman in a long red coat walks away from camera down a narrow cobbled street. Tracking shot following from behind. Evening, warm shop lights. Steady pace." Walking away is far more reliable than walking towards camera.
Try it
A barista in a linen apron pours milk into a cup, a slow steady stream forming a pattern. Static close-up on the cup and hands. Small cafe, morning window light, warm wood tones. Slow, calm, natural motion.
Run it, then change only the camera line to "slow push in towards the cup" and compare. Then try removing "slow, calm, natural motion" to see how much the pace sentence was doing. Keep the version that holds the hands steady; hands are the first thing to break.