MiniMax H3 AI Video Generator (Hailuo 3.0)

Create 4–15-second videos with MiniMax H3, also known as Hailuo 3.0, from text, a first and optional last frame, or image, video, and audio references. Choose 768p or 2K output with native generated audio, and see the exact credit estimate before you generate.

From 12 credits
768p or 2K
Text, frames, or references
MiniMax · MiniMax H3 docs ↗

MiniMax H3 Video Examples

Clay fox leaping a lava canyon between two keyframes, generated by MiniMax H3

First and last frame

Prompt

Claymation. Start on @Image1 and finish on @Image2. A fox sprints to the edge of a cliff and launches without hesitation, making a dramatic heroic leap in slow motion over an immense lava canyon. Midair, the camera races beneath the fox’s belly in a bold dynamic move, revealing the terrifying depth of the chasm and the fully extended motion of its clay body.

Key Features of MiniMax H3

MiniMax H3 unifies multimodal understanding, generation, and instruction-based editing in one model—built for flexible creation across film, advertising, product, and social content.

Multimodal reference

Combine up to 9 images, 3 videos, and 3 audio clips—12 files total. Tag each file so it locks identity, motion, camera, or voice in one take.

Cafe singer filmed with Hitchcock camera language, generated by MiniMax H3

How to Use MiniMax H3

1
1. Start from text, frames, or references
Use text-to-video for a new scene, add an owned first frame with an optional end frame for a controlled transition, or switch to multimodal mode for up to 9 images, 3 videos, and 3 audio clips—12 files total.
Use text-to-video for a new scene, add an owned first frame with an optional end frame for a controlled transition, or switch to multimodal mode for up to 9 images, 3 videos, and 3 audio clips—12 files total.
2
2. Direct motion, camera, and sound
Describe subject, action, camera movement, dialogue, ambience, and timing. In multimodal mode, use the composer’s tags so each file stays assigned. To edit an owned clip, name one change and what must stay the same.
Describe subject, action, camera movement, dialogue, ambience, and timing. In multimodal mode, use the composer’s tags so each file stays assigned. To edit an owned clip, name one change and what must stay the same.
3
3. Generate, review, then iterate
Select 4–15 seconds and 768p or 2K, confirm the complete variable credit estimate, then generate. Review identity, readable text, motion continuity, dialogue, and lip sync before publishing or iterating.
Select 4–15 seconds and 768p or 2K, confirm the complete variable credit estimate, then generate. Review identity, readable text, motion continuity, dialogue, and lip sync before publishing or iterating.

Create with MiniMax H3 ↑

MiniMax H3 Pricing

From 12 credits per 4-second 768p clip, 15 credits for the default 5-second 768p take, up to 60 for 15-second 2K text-to-video. Fully loaded reference-led requests can reach 128 credits. Duration, 768p or 2K, and reference-video seconds change the quote. Confirm the exact reservation before you generate.

Compare MiniMax H3How MiniMax H3 compares to other models

MiniMax H3

This model

Multimodal reference, first/last frames, and instruction-led edits

4–15s12–128 credits

MiniMax H3 Max →MiniMax

Fast 5–15s drafts with first/last frames or a 9/3/3 reference pack

5–15s18–321 credits

Seedance 2.5 →ByteDance

4–30s one-takes, packs up to 50 files, and clay-render blocking

4–30s38–1320 credits

Attribute	This modelMiniMax H3	MiniMaxMiniMax H3 Max →	ByteDanceSeedance 2.5 →	KuaishouKling 3.0 →
Best for	Multimodal reference, first/last frames, and instruction-led edits	Fast 5–15s drafts with first/last frames or a 9/3/3 reference pack	4–30s one-takes, packs up to 50 files, and clay-render blocking	5/10/15s cinematic shots from text or an owned frame
Inputs	Text · First/last frame · Up to 9 images · 3 videos · 3 audio	Text · First/last frame · Up to 9 images · 3 videos · 3 audio	Text · Up to 30 images · 10 videos · 10 audio	Text · Image + end frame
Output	768p–2K video + audio	480p–1080p video + audio	480p–1080p video + audio	720p–4K video + audio
Duration	4–15s	5–15s	4–30s	5–15s
Native audio	Yes	Yes	Yes	Yes
Credits	12–128 credits	18–321 credits	38–1320 credits	39–385 credits

Stay on MiniMax H3 for multimodal packs, first/last frames, and instruction-led edits at 768p or 2K. Switch to MiniMax H3 Max for a faster 5–15 second 480p–1080p draft, Seedance 2.5 for a 4–30 second one-take, or Kling 3.0 for 5/10/15s cinematic shots.

MiniMax H3MiniMax H3 MaxSeedance 2.5Kling 3.0

MiniMax H3 FAQs

What is MiniMax H3 (Hailuo 3.0)?

MiniMax H3, also known as Hailuo 3.0, is MiniMax’s general-purpose multimodal video model. On MixVio it creates 4–15-second videos from text, a first frame and optional last frame, or image, video, and audio references. MiniMax H3 announcement ↗

Is MiniMax H3 the same as Hailuo 3.0?

Yes. MiniMax H3 is the official name; Hailuo 3, Hailuo 3.0, and Hailuo H3 all refer to the same MiniMax video model. MixVio lists it as MiniMax H3 (Hailuo 3.0).

What inputs and modes does MixVio support?

Use text-to-video for a new scene, first-frame image-to-video with an optional last frame, or multimodal reference-to-video for mixed files. First/end-frame mode and multimodal mode are mutually exclusive. In multimodal mode, mention each file with @ImageN, @VideoN, or @AudioN so roles stay assigned. Prompts can be up to 7,000 characters.

How many references can I add?

Up to 9 images (≤30 MB each), 3 videos (≤50 MB each), and 3 audio clips (≤15 MB each), with no more than 12 files total. Still images need 256–5,760 px on each side and a 2:5–5:2 crop; oversized photos are fitted automatically. Reference videos also use 256–5,760 px per side—not a 480p–720p short-edge rule. Reference video and audio may total up to 15 seconds each; audio cannot be the only reference type.

Does MiniMax H3 generate native audio?

Yes. Every MiniMax H3 generation returns video with native stereo audio in the same pass. Describe dialogue, ambience, sound effects, or music in the brief. A reference recording can guide voice or atmosphere when it is paired with an image or video. Review speech, mix, and lip sync before publishing.

Can I edit a video I already have?

Yes, as a single-shot instruction edit—not a timeline editor. Upload an owned clip in multimodal reference mode, name one change and what must stay the same, then generate. MixVio returns one new private 4–15 second video per run. First/last-frame mode cannot be combined with this path, and there is no multi-turn session editor on MixVio.

Which durations, resolutions, and ratios are supported?

MixVio supports 4–15 second output at 768p or 2K, in whole seconds. The duration you select is never lengthened or changed without confirmation. Text-to-video supports 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. First-frame mode follows the source; multimodal mode also supports adaptive framing.

How long does MiniMax H3 take?

On MixVio’s 2026-09-01 Denver-day sample, typical MiniMax H3 completion was about 90 seconds while the generator stayed open. Keep this page open for the 1–2 minute wait. If you leave, the clip still appears in History when it is ready.

How many credits does MiniMax H3 use?

MiniMax H3 is available on the Free plan when your balance covers the quote. Text-to-video runs reserve from 12 to 60 credits. Fully loaded reference-led requests can reach 128 credits because reference-video seconds and images after the first five are also billable. The generator shows the exact reservation before you submit. Failed runs cost no credits.

Is MiniMax H3 free?

It is available on the Free plan, but generation is not zero-credit. The exact quote depends on duration, resolution, and references; failed runs cost no credits. Free delivery limits still apply.

How does MiniMax H3 compare with Seedance 2.5, Seedance 2.0, and Kling 3.0?

Stay on MiniMax H3 for multimodal packs, first/last frames, and instruction-led edits at 768p or 2K. Seedance 2.5 is the 4–30 second one-take route with a larger reference pack and 480p–1080p. Seedance 2.0 stays on 4–15 seconds with selectable 480p–4K. Kling 3.0 Standard offers 5/10/15 second cinematic shots. Compare live credits, inputs, duration, and audio in the comparison table before generating.

Can I use MiniMax H3 videos commercially?

Yes—paid MixVio generations can be used commercially, subject to the Terms of Service and current MiniMax / serving provider terms, when you have rights to every prompt, reference, likeness, brand asset, voice, and sound direction. Completed outputs are archived privately to your MixVio workspace. Prompts and references pass MixVio moderation and Provider safety checks; some requests may be rejected. Review identity, readable text, logos, dialogue, lip sync, and audio before publishing. MixVio does not guarantee exclusivity or trademark clearance.

Can I use real faces or celebrity likenesses?

Only use faces, voices, and likenesses you own or are authorized to use. Public figures, deceptive impersonation, deepfake-style face swaps, minors, and other prohibited content can be blocked by MixVio moderation or Provider safety checks. When a request is allowed, still review identity and consent before any commercial or public use.

Direct text, frames, and references with MiniMax H3.
Private. Variable cost shown first.

Start from text, frames, or an owned reference pack, then generate 768p or 2K video with native audio—MiniMax H3 selected.