This is the long version: what the current Hailuo model can do, where the ceilings are, what everything costs, and how the pieces fit into an actual workflow. If you want the short introduction instead, start with What Is Hailuo?.
Hailuo 03, released in July 2026, is the current default model on HailuoLabs. It replaced Hailuo 2.3 at no extra cost — same plans, same prices, same allowances — and the older models stayed selectable.
What changed in the current model
| Hailuo 2.3 | Hailuo 03 | |
|---|---|---|
| Max duration | 15 seconds | 30 seconds |
| References | Up to 12 (9 images, 3 clips, 3 audio) | Up to 50, any mix |
| Fixing a bad detail | Regenerate the whole clip | Frame-local redraw |
| Audio and dialogue | Stereo, generated with the picture | Stereo, generated with the picture |
| Max resolution | 4K | 4K |
Two of those rows change how you work rather than how the output looks, and they are the reason the update matters.
Thirty seconds in one generation
The 15-second ceiling on earlier models quietly shaped everything people made with them. A 30-second ad became three clips and a stitch, and every stitch was a place for the character, the colour, or the light to drift.
At 30 seconds, a complete deliverable fits in one pass:
- A full-length social ad with a hook, a demonstration, and an end card.
- A verse of a music video cut against an attached audio reference.
- A narrative scene with a beginning, a turn, and an ending.
- A product walkthrough that gets through more than one feature.
Shorter is still often better. Thirty seconds is a ceiling, not a target, and a six-second clip that lands beats a thirty-second clip that wanders.
Fifty references, mixed freely
The reference ceiling went from 12 to 50, and the rigid split — nine images, three clips, three audio — went away. You can now bring 50 images, or 30 images and 15 clips and 5 audio tracks, or any other combination.
At that scale references stop being hints and start being a brief:
- Character sheets. A face from several angles, in several lighting conditions, so identity holds across shots.
- Brand kits. Product shots, packaging, palette swatches, typography boards.
- Motion references. Clips showing the camera move or the physical action you want echoed.
- Audio. The actual track, so cuts land on the beat instead of near it.
Frame-local redraw
Previously, one wrong hand meant regenerating the clip and hoping the rest came back as good. Frame-local redraw changes that: circle the region that is wrong, describe the fix, and only that area is regenerated. Everything you already approved stays byte-for-byte as it was.
Practically, this is what makes iteration affordable. It is worth pushing a generation you are 90% happy with over the line rather than rerolling for a clean take, because a reroll puts the 90% you liked back at risk.
Native audio
Score, ambient effects, and character dialogue are generated in the same pass as the picture and synced to the action, in stereo. That has consequences for how you prompt:
- Audio direction belongs in the prompt. Name the ambience, name the score, or say "no music".
- Dialogue goes in quotes with an attribution, and gets lip-synced.
- An attached audio reference gives the model the timing to cut against.
Formats and resolutions
| Setting | Available options |
|---|---|
| Aspect ratio | Auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 |
| Resolution | 768p or 2K in the generator; up to 4K output |
| Duration | 4–15 seconds on 2.3, 2.3 Fast and 02; up to 30 seconds on Hailuo 03 |
Pick the aspect ratio before you write the prompt, not after. A shot composed for 16:9 does not survive a crop to 9:16, and the model frames differently depending on which one you have chosen.
Plans, credits and limits
| Plan | Price | Credits / month | Roughly |
|---|---|---|---|
| Free | $0 | 160 | 5 videos or 10 images, 480p, watermarked |
| Creator | $14 | 2,800 | 96 videos or 230 images |
| Studio | $49 | 10,000 | 344 videos or 830 images |
| Agency | $99 | 22,000 | 758 videos or 1,830 images |
One-time credit packs are available alongside the plans: $50 for 10,000 credits, or $100 for 20,500 at the better rate. Plan credits reset each billing cycle; purchased packs never expire.
All paid plans include a commercial licence and the media gallery. The free plan is watermarked, capped at 480p, and does not carry a commercial licence.
A working process
- Draft cheap. Iterate at 768p on a fast model until the composition and the action are right.
- Lock references. Once the shot works, attach the images and audio that must stay consistent.
- Run at quality. Rerun the winning prompt on Hailuo 03 at 2K.
- Patch, do not reroll. Fix stray details with frame-local redraw.
- Export and archive. Download before the 30-day window closes.
The limits worth knowing
- Thirty seconds is the maximum clip length; longer pieces still need an edit.
- Hailuo is proprietary. There are no open weights to self-host.
- The free tier is for evaluation — watermarked, 480p, no commercial licence.
- Public leaderboard rankings shift between releases. Judge models on your own shots.
FAQ
Do I have to pay extra for the newest model?
No. Hailuo 03 replaced 2.3 as the default at no extra cost — same plans, same prices, same allowances.
Can I still use the older models?
Yes. Hailuo 2.3, 2.3 Fast, and 02 all remain selectable in the model picker.
How many references can I attach?
Up to 50 in a single generation on Hailuo 03, in any mix of images, video clips, and audio. Older models cap at 12.
What is the maximum resolution?
The generator offers 768p and 2K, with output up to 4K.
Do credits roll over?
Monthly plan credits reset each cycle. One-time credit packs never expire.
How long are my generations kept?
Thirty days, after which they are deleted from your library automatically.
Is there a commercial licence?
Yes, on all paid plans. The free plan does not include one.
