Prompting Hailuo 2.3
MiniMax Hailuo 2.3 shines at human motion, stylized (anime/ink-wash/game-CG) video, and e-commerce object motion — and it's cheap enough to iterate a lot. This guide shows how to structure prompts. Model overview: Hailuo API guide.
The Camera–Character–Reaction Structure
Hailuo responds well to a three-part structure:
- Camera: how the shot moves (slow tracking, push-in, static).
- Character/Subject: who/what and the specific action.
- Reaction: the resulting motion, expression, or physics.
Camera: slow push-in;
Character: an anime girl in a yellow raincoat walks through neon Tokyo rain;
Reaction: hair and coat sway, puddle ripples, expressive smile
Nail the Style
Hailuo commits to non-photoreal styles that realism-biased models render awkwardly. Name the style explicitly: "anime", "ink-wash painting", "game CG", "cel-shaded". This is where Hailuo beats general models.
E-commerce Object Motion
For product spins, be precise about the motion and what stays readable:
Camera: static;
Subject: a glass perfume bottle rotates 360° on a marble surface;
Reaction: label stays facing camera, soft studio reflections
Generate a Clip
curl -X POST https://api.linkmodel.ai/api/v1/video-generation \
-H "Authorization: Bearer $LINKMODEL_API_KEY" -H "Content-Type: application/json" \
-d '{ "model": "MiniMax-Hailuo-2.3", "prompt": "<your Camera-Character-Reaction prompt>", "resolution": "1080P", "duration": 6 }'Submit, get a task_id and price, poll for the video. Note: 1080p at 6s or 768p at 10s; no native 4K (use Kling V3 for 4K). Confirm parameters in the docs.
Cost-Smart Iteration
Hailuo has a Fast variant (50% cheaper) — draft on Fast, render finals on standard. It's cheap ($0.08/s), so generate several variations and pick the best. More cost tactics in how to reduce AI API costs.
Tips
- One clear action per clip.
- Name the style for anime/illustration work.
- Describe physics ("cloth sways", "ripples") for believable motion.
- Need synced audio? Hailuo focuses on visuals — see AI video with audio for single-pass audio models, and Hailuo vs Seedance.
Bottom Line
Use Camera–Character–Reaction, name the style, be precise about motion, and iterate cheaply on the Fast tier. More models in best AI video generation APIs.
Start free with a $1 credit and test your prompts.