What’s new
Source material here is limited: no dedicated MiniMax announcement for Hailuo 2.3 turned up, only the integrator's API reference — so this list covers confirmed capabilities rather than a full comparison with the prior version.
- 15 built-in camera-move commands: bracket a camera keyword into the prompt (e.g. "[push in]") and the docs confirm 15 supported movement commands.
- Prompt auto-optimization is on by default: the API rewrites your prompt for you unless you turn that off and write it exactly as-is.
- A dedicated Fast variant exists: MiniMax-Hailuo-2.3-Fast, for lower latency, but it requires a first-frame image and can't run as pure text-to-video.
What it’s good for
- Short clips where you want precise camera control via bracketed movement tags.
- Turning an existing reference image into a moving clip quickly (especially with the Fast variant).
- Higher-volume work where resolution requirements are modest and turnaround matters.
Prompting tips
- Use bracketed camera tags, like "[push in]" — the docs confirm 15 supported movement commands.
- 1080p only supports a 6-second duration; for 10 seconds, use 768p instead.
- Prompt auto-optimization is on by default — turn it off if you want the prompt executed exactly as written.
- Use the Fast variant for quicker turnaround, but note it requires a first-frame image — it can't run as pure text-to-video.
- Prompts are capped at 2,000 characters, tighter than some comparable models, so keep longer descriptions concise.
768p, 10 seconds.
[push in] A vintage motorcycle rides into frame, rider in a black leather jacket.
[orbit] The camera slowly circles the motorcycle, showing off its details and the sunset reflecting off the chrome.
Known limits
- At 1080p, duration is limited to 6 seconds; for 10 seconds you need to drop to 768p.
- The Fast variant requires a first-frame image — it doesn't support pure text-to-video.
- Prompts max out at 2,000 characters.
Using it in Nomi
On the generation canvas, add a video card, pick “Hailuo 2.3” in the model picker and choose a mode (Text to video, First frame to video). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
