Hailuo AI: MiniMax's Video Models and How to Use Them

Hailuo AI is the video model line built by MiniMax, and the name trips people up because the company and the product are different words. Seven Hailuo builds run on the FLORA canvas across three generations, from a two hundred second run to a cinematic Pro build.

Want to try it? All three Hailuo generations run on the FLORA canvas, so you can send one prompt through each and watch where the extra cost actually goes. Start free on FLORA.

What is Hailuo AI?

Hailuo is the name MiniMax gives its video generation models. MiniMax is a Chinese AI company that also builds large language models, so if you search the company name you will mostly find those. Hailuo is the video half, and it is the name worth knowing when you are picking a model to animate something.

The line has moved quickly. The first Hailuo build established the basics, Hailuo 02 raised the quality, and Hailuo 2.3 is the current generation, which MiniMax tuned specifically around physical action, stylization and the small expressions that make a face read as a performance. On FLORA all three are available, which makes this one of the few model families where you can hold a prompt still and watch a generation improve.

Quick facts

• What runs on FLORA: seven builds across three generations, Hailuo, Hailuo 02 Pro and Hailuo 2.3 Pro.
• Modes: text to video and image to video on all three. Hailuo 02 Pro also takes a first and last frame.
• Duration: 6 or 10 seconds.
• Resolution: 768p or 1080p, with 1080p limited to the 6 second length.
• Frame rate: 24 fps.
• Cost on FLORA: from about 0.23 US dollars on the first generation to about 0.58 on 2.3 Pro.
• Speed: 210 to 420 seconds depending on the build.
• Provider: MiniMax. Proprietary, so there is no self hosted option.

What it is great for

• Shots that turn on a physical action, like something being picked up, poured or set down, which is where 2.3 put its improvements.
• Character work at close range, where micro-expressions carry the shot.
• Stylized and animated looks, which the line handles as readily as photoreal ones.
• Bridging two stills through the first and last frame mode on Hailuo 02 Pro, when you know where a shot starts and ends.

The three generations compared

This is the decision most people arrive at this page to make. The builds differ in cost by about a factor of two and in speed by a factor of two in the opposite direction, so the cheap one is also the fast one.

Example outputs

Hailuo, the first generation. A 5.6 second clip at 1280 by 720, about 3.5 minutes to generate.

Hailuo 2.3 Pro, same prompt. A 5.9 second clip at 1920 by 1080, about 5 minutes to generate. The newer build returned double the resolution.

Copy and paste prompts

A paper boat drifting along a rain gutter after a storm, low camera close to the water, afternoon light

Close on a hand setting a ceramic cup down on a wooden table, slight ripple in the liquid, 6 seconds

Stylized animation: a kite lifts off in three beats, rises past the frame edge, cut to sky

Parameters

Builds on the canvas

Prompt tips

• Lead with the physical action. Hailuo 2.3 was tuned on exactly this, so a prompt built around one concrete movement gets more out of it than a prompt built around a mood.
• Pick 768p while you are exploring. You get the 10 second length there, and 1080p caps you at 6 seconds anyway.
• Keep close shots close. The micro-expression work shows at a tight framing and is largely wasted in a wide.
• Say the style outright if you want one. The line handles stylized and animated looks well, but it defaults toward photoreal unless told otherwise.
• Use the first and last frame mode when you already know both ends of a shot. Describing the middle is easier than describing the whole thing.

Running Hailuo AI in FLORA

On the FLORA canvas each Hailuo build sits as its own block, so you can wire the same prompt into two of them and compare what comes back without retyping anything. That is the fastest way to answer the cost question for your own work, because the gap between the cheap build and the Pro build matters more on some shots than others.

The techniques handle the steps either side. Start a clip from a description with text to video AI, or animate something you already have with image to video AI and the AI image to video generator. If the still needs work first, the background remover and the image upscaler are on the same board.

Models are interchangeable here, so put Hailuo against the others on a shot you actually care about. Seedance handles long single takes, Kling is strong on physical plausibility, and Veo generates sound with the picture. Our ranking of the best AI video generators covers the wider field.

Hailuo AI FAQs

Who makes Hailuo AI?

MiniMax, a Chinese AI company that also builds large language models. Hailuo is the name of its video model line, which is why searching the company name turns up mostly language model results.

What is the difference between Hailuo, Hailuo 02 and Hailuo 2.3?

They are three generations of the same line. The first build is the cheapest and fastest, Hailuo 02 Pro raised fidelity and added a first and last frame mode, and Hailuo 2.3 Pro is the current generation, tuned around physical action, stylization and facial micro-expressions.

How long can a Hailuo clip be?

6 or 10 seconds. The 10 second length is available at 768p, and 1080p caps at 6 seconds, so there is a real trade between length and resolution, and no single best setting.

How much does Hailuo AI cost on FLORA?

From about 0.23 US dollars a run on the first generation to about 0.58 on Hailuo 2.3 Pro. You can try any of them on FLORA's free tier before committing budget.

Is Hailuo AI better than Kling or Seedance?

Each one has a different strength. Hailuo 2.3 is tuned hardest on physical action and close-range expression, Kling holds physical plausibility well, and Seedance manages long single takes. All three run on the FLORA canvas, so put your own prompt through each and judge the shot in front of you.