The gap between "I should try an AI video tool" and "I have a clip I can actually use" is usually about five minutes. Here is the whole first run, in order, with the decisions that matter flagged along the way.
Step 1: Create a free account
Sign up and you land in the prompt box with 160 free credits on the clock — roughly five videos. No card, no waitlist, no invite code, and nothing to install. Hailuo 03 is already selected as the default model.
Free-plan output is 480p with a watermark. That is fine for learning the tool; upgrade when you want the clean export.
Step 2: Write a prompt that directs the shot
This is the step that decides whether the other four are worth doing. The instinct is to describe a scene the way you would describe a photograph. Direct it instead, the way you would brief a camera operator.
A prompt that works has most of these parts:
- Subject — who or what is in frame, described specifically enough to stay consistent.
- Action — one clear beat. Not three.
- Setting — where, and what time of day.
- Camera — the shot size and the move. "Slow push in on a medium close-up" beats "cinematic".
- Light and style — the look. Hard sun, soft overcast, neon spill, film stock.
- Sound — what you want to hear, including any dialogue in quotes.
- Constraints — what to keep out.
Weak prompt:
- A beautiful cinematic video of a chef in a kitchen, amazing quality, 4k
Stronger prompt:
- Medium close-up of a chef plating a dish under warm overhead pendant lights in a dark restaurant kitchen. She sets down the last garnish and looks up. Camera pushes in slowly. Shallow depth of field, steam catching the light. Ambient kitchen clatter, low room tone. No music, no on-screen text.
The second one gives the model a shot size, a move, a light source, one action, and an audio direction. The first gives it adjectives.
Step 3: Add references
References are how you control the things words are bad at — a specific face, a product's exact shape, a brand palette, the tempo of a track.
Hailuo 03 accepts up to 50 references in one generation, and you can mix types however you like: images, video clips, and audio in any combination. Older models cap at 12 (nine images, three clips, three audio).
- Images pin down look: characters, products, locations, colour.
- Video clips pin down motion: a camera move or an action you want echoed.
- Audio pins down rhythm: attach the track and the cuts land against it.
A useful rule: words set the look, references set the specifics. If you have written the same descriptive sentence three times and the face still changes between generations, stop writing and attach a photo.
Step 4: Choose model, format and length
| Setting | Options | How to choose |
|---|---|---|
| Model | Hailuo 03, 2.3, 2.3 Fast, 02 | Start with 03. Switch to 2.3 Fast when you are iterating and want speed over fidelity |
| Aspect ratio | Auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | 9:16 for TikTok, Reels and Shorts. 16:9 for YouTube and web. 1:1 for feed ads |
| Resolution | 768p or 2K | Draft at 768p, final at 2K. Rendering drafts at full resolution burns credits on takes you will throw away |
| Duration | 4–15s on older models, up to 30s on Hailuo 03 | Shortest length that holds the beat. Longer is not better and costs more |
Step 5: Generate, fix, export
Hit generate and wait. When the clip comes back, watch it once at speed and once looking for the specific thing you care about.
If one detail is wrong — a hand, a label, a background object — you do not have to reroll. On Hailuo 03 you can circle that region and redraw just it, and the rest of the clip stays exactly as you approved it. This is the feature that turns generation from a slot machine into an edit.
When the take is right, export. Paid plans export without a watermark and include a commercial licence.
When the first try does not work
- The face or product changes between shots. Attach an image reference instead of describing it again.
- Motion is mushy or the subject drifts. Your prompt probably has more than one action in it. Cut it to one beat.
- The clip ignores half your prompt. Front-load what matters. Subject and action first, style last.
- The audio is wrong. Say what you want to hear explicitly, and say what you do not — "no music" is a valid instruction.
- It looks like stock footage. You gave it adjectives instead of a camera. Add a shot size and a move.
FAQ
Do I need to install anything?
No. Hailuo runs entirely in the browser.
How many free videos do I get?
The free plan includes 160 credits a month, which is around five videos, at 480p with a watermark.
Can I use the videos commercially?
Paid plans include a commercial licence. The free plan does not, and its exports are watermarked.
Which model should I start with?
Hailuo 03 — it is the default, and it has the longest clips, the highest reference ceiling, and frame-local redraw.
How long can my video be?
Up to 30 seconds on Hailuo 03, and 4–15 seconds on the older models.
Can I add my own music?
You can attach an audio track as a reference, and the generation will work with it rather than against it.
