This page tracks what has shipped on HailuoLabs. The most recent release is at the top.
Hailuo 03 is live — July 2026
Hailuo 03 is now the default model in the prompt box, replacing Hailuo 2.3 at no extra cost — same plans, same prices, same allowances. No waitlist, no staged rollout, and nothing to opt into: existing accounts got it on launch day, and the free plan included.
Three changes account for most of the difference.
Clips go to 30 seconds
Double the previous 15-second ceiling. A complete 30-second ad, a full verse of a music video, or a scene with a beginning and an ending now fits inside a single generation — no stitching, and no drift between joined clips.
Up to 50 references, mixed freely
The old structure allowed 12 references in fixed slots: nine images, three video clips, three audio tracks. Hailuo 03 takes up to 50 in any combination. Fifty images if that is what you need, or thirty images with fifteen clips and five audio tracks.
That is enough to hand the model an entire character sheet, a brand kit, and a music bed at once, rather than choosing between them.
Frame-local redraw
The change that most affects how the tool feels day to day. When one detail in an otherwise good clip is wrong, you circle it and redraw just that region. Everything you already approved is untouched.
Before this, a single bad hand meant regenerating the whole clip and gambling the take you liked. Now iteration is additive rather than a reroll.
What stayed the same
Resolution still tops out at 4K. Native stereo audio — music, ambient effects, and lip-synced dialogue generated in the same pass as the picture — carried over unchanged. Aspect ratios, plans, credits, and the commercial licence on paid tiers are all as they were.
Hailuo 2.3 — February 2026
The previous flagship. At release it topped the independent Artificial Analysis arena for both text-to-video and image-to-video.
It introduced the native audio pipeline that the current model still uses: stereo music, ambient sound, and character dialogue generated alongside the video and synced to the action. Clips ran 4–15 seconds at up to 4K, across 16:9, 9:16, 4:3, 3:4, 21:9, and 1:1, with up to 12 references in fixed slots.
A speed-optimised variant, Hailuo 2.3 Fast, is also available for quicker turnarounds.
Earlier: Hailuo 02
The generation before 2.3, and still selectable in the model picker. It remains the model behind the 512p video output on the free plan.
What to do with a new model
- Rerun your best prompt unchanged. The clearest read on what changed is your own shot, not a demo reel.
- Retry the thing that used to fail. Consistency problems and fiddly details are where new releases tend to move most.
- Raise your reference count before rewriting prompts. With a 50-reference ceiling, drift is usually solved by attaching more, not by writing more.
- Keep the old model around. Nothing forces you off a workflow that already works.
FAQ
What is the newest Hailuo model?
Hailuo 03, released in July 2026. It is the default in the prompt box.
Does the new model cost extra?
No. It replaced Hailuo 2.3 as the default at no extra cost, on the same plans and allowances.
Did I need to do anything to get it?
No. It rolled out to all accounts, including free plans, with no waitlist or opt-in.
Are the older models gone?
No. Hailuo 2.3, 2.3 Fast, and 02 are all still selectable.
What is the biggest practical change?
Frame-local redraw. Fixing one detail no longer means regenerating the clip.
