Hailuo vs MiniMax H3: the same vendor, two opposite camera grammars

MiniMax ships two video lines on LUVI, and they want the camera written in opposite ways. On Hailuo it is a bracketed command from a closed list—[Push in], [Tracking shot], [Static shot]. On MiniMax H3 those brackets do nothing at all, and the camera has to be described in prose. The practical consequence is the reason for this post: a prompt that works on one line loses its camera move on the other without any error to tell you.

What exactly changes?

The instruction becomes a sentence, and it gains two dials.

• Hailuo: A man picks up a book [Pedestal up], then reads [Static shot]. Fifteen commands, exact spelling, placed inline where the move happens. Up to three inside one pair of brackets run at once.
• H3: The camera pushes in with small amplitude at slow speed toward the folded letter in her hands. Twelve moves, written out, with four modifier literals—with small amplitude, with large amplitude, at slow speed, at fast speed. Medium amplitude and normal speed are expressed by leaving them out.

H3 also spends a cut rather than a comma on a second move: one shot carries one move, one visible action and one end frame, and a change of viewpoint is written as a new shot with a timestamp. Hailuo's rule is looser—separate brackets in text order, joined by "then".

What else does not transfer?

Four things, and the audio one is the largest.

• Sound. Hailuo renders picture only: no dialogue, no effects, no music, so every word about audio is wasted. H3 generates audio always and wants two fields filled in—one paragraph of soundscape and one of score—because an omitted field is filled with invented sound rather than silence.
• Prompt shape. A Hailuo prompt is one clear paragraph under 2,000 characters. An H3 prompt is a fielded document with lowercase field names and blank lines between them.
• Dialogue languages. H3's stable set covers eleven languages; Turkish is outside it, and the safe route is to write the shot with no dialogue block and lay the voice in afterwards. Hailuo has no dialogue at all, so the question does not arise.
• Proper nouns. On H3, character names, film titles and celebrities trip input moderation and block the request outright. Hailuo also rejects a real person as a subject, but the failure is softer—describe the look instead, on both.

Which parameters differ, and what do they cost?

Both are per-second, and both hide a default worth checking.

• Hailuo-2.3 Standard lets you pick six or ten seconds; the Pro tier has no duration control and renders six. A six-second clip quoted 560 credits on Standard and 980 on Pro, with 380 on Fast.
• H3 runs from 4 to 15 seconds, with three resolutions written in capitals—480P, 768P, 2K—and it defaults to 2K. At eight seconds it quoted 608 credits at 480P and 1,280 at 768P.

Two defaults deserve a deliberate choice. H3's aspect ratio starts at 1:1, not 16:9. And its resolution enum is case-exact: 768P with a capital P. A value in the wrong case is dropped rather than corrected, so the model falls back to its own default and you pay for a tier you did not pick.

So which line for which job?

• Picture-only work where physics has to look right—an object falling, a hand turning something over, a vehicle moving through a frame—goes to Hailuo, and the bracketed commands give you tighter control of the move than prose does.
• Anything that needs its own sound, dialogue included, goes to H3, because Hailuo cannot make any.
• A single clean move on a tight budget is Hailuo Fast territory at 380 credits for six seconds.
• A longer beat, or several beats with cuts, is H3: fifteen seconds against ten, and cuts are part of its grammar rather than something to avoid.

How we checked

• Models: Hailuo-2.3 t2v Standard and t2v Pro, and MiniMax H3 Text-to-Video.
• Settings: six-second Hailuo clips; eight-second H3 clips at 480P and 768P, 16:9.
• Date and what was read: September 22, 2026—live credit quotes and the models' parameter contracts in LUVI, LUVI's own prompting guides for both families, and MiniMax's text-to-video API reference for the Hailuo command list.
• Results: 560 and 980 credits for six seconds on Hailuo-2.3 Standard and Pro, 380 on Fast; 608 and 1,280 credits for eight seconds on H3 at 480P and 768P. No outputs were generated for this post, so it makes no claim about output quality.
• Drawbacks: the claim that brackets do nothing on H3 comes from LUVI's prompting guide for the family rather than a published test by the vendor.

In LUVI

The feather button beside the prompt box is the safety net here: it applies the prompting guide of whichever model is selected, so a bracket that wandered from Hailuo into an H3 prompt gets rewritten as prose rather than rendered as nothing. Before you generate on H3, check the aspect ratio and the resolution in the controls—both defaults are ones people don't expect. The help center's list of which models generate sound with the video settles the other half of the choice.

New accounts start with 200 credits, and verifying a phone and a card adds up to 1,500 more. You can create a LUVI account and feel the difference between the two grammars on one shot.

Sources

• Video generation (text-to-video) API reference — MiniMax, official API documentation. Accessed September 22, 2026.