MiniMax Hailuo
Chinese AI lab behind the H3 video model: cheap per-second 2K clips with native audio, open weights
MiniMax makes the H3 family of video models (H3, H3-Max, H3-Regeneration), launched in summer 2026, plus the older Hailuo 2.3 line, speech, music and image models. H3 generates 4 to 15 second clips in 768P or 2K, through a published per-second API. The H3 weights are also on Hugging Face for local 768P inference under a restrictive community license.
- 18 creators using
- 5 news items tracked
- 2024 first launched
- Freemium pricing
- Easy learning curve
What MiniMax Hailuo can do
- Infinite canvas Not confirmed
- AI agent Yes
- Node / flow builder No
- Public API Yes
- MCP server Yes
16 of 26 capabilities confirmed
| Capability | Supported | Details |
|---|---|---|
| Text to image | Yes | image-01 at $0.0035 per image |
| Image editing | Not confirmed | |
| Text to video | Yes | H3 and H3-Max, 4 to 15 seconds |
| Image to video | Yes | First and last frame, 0 to 2 images |
| Video to video | Yes | H3 motion transfer and reference-based editing; input video $0.13/s at 2K |
| Native audio in video | Yes | H3 native stereo audio |
| Lip-sync | Not confirmed | |
| Voice / TTS | Yes | Speech 2.8 TTS and voice cloning ($1.5 per cloned voice) |
| Music & SFX | Partial | Music-3.0 open weights; paid music API closed to new users from 20 Aug 2026 |
| Upscaling | Partial | H3-Regeneration upgrades 768P to 2K; not a general-purpose upscaler |
| Character consistency | Yes | Reference generation with up to 9 images, 3 video and 3 audio references |
| Camera control | Partial | Prompt tokens [pan], [zoom], [static]; no shot-by-shot storyboard tool |
| Custom training | Not confirmed | |
| Third-party models | No |
| Capability | Supported | Details |
|---|---|---|
| Infinite canvas | Not confirmed | |
| Node / flow builder | No | Official pages describe agent-driven workflows and no visual node builder |
| AI agent | Yes | MiniMax Design: agent platform for ads and e-commerce content; MiniMax Code: coding agent harness |
| Chat assistant | Not confirmed | |
| Team collaboration | Not confirmed |
| Capability | Supported | Details |
|---|---|---|
| Public API | Yes | Public pricing and API docs at platform.minimax.io |
| MCP server | Yes | Official MiniMax-MCP server (generate_video, async mode) |
| Mobile app | Not confirmed | |
| Desktop / local | Yes | H3 open weights run locally at 768P; MiniMax Code has a desktop download |
| Open source | Partial | H3 weights open under the MiniMax H3 Community License, which is not an OSI open-source license |
| Capability | Supported | Details |
|---|---|---|
| Free tier | Partial | |
| Commercial use | Not confirmed |
Yes Partial No Not confirmed by the vendor
What makes MiniMax Hailuo different
- 01
Published per-second API pricing you can budget against: $0.08/s at 768P, $0.13/s at 2K, $0.05/s for 480P H3-Max.
- 02
H3-Regeneration takes a 768P render to 2K at $0.05/s, so you can generate cheap variants and only upscale the picks.
- 03
Open weights on Hugging Face for local 768P inference, which almost no closed competitor offers.
- 04
Multi-reference input: up to 9 images, 3 video clips and 3 audio clips per request to hold a character, product or voice across shots.
Latest from MiniMax Hailuo
- May 26 MiniMax Music 2.6 Updates Multi-Voice Synthesis and Emotional Narrative Control MiniMax Music 2.6 introduces advanced multi-character vocal synthesis and improved emotional range for AI-generated music and soundtracks.
- May 26 MiniMax Speech 2.8 Introduces High-Fidelity Human Emotion and Nuance to AI Voice Cloning The MiniMax Speech 2.8 update brings advanced emotional range and natural human prosody to its voice generation model for more realistic narration.
- May 26 Hailuo 02 Launches with NCR Architecture and Native 1080p Generation MiniMax releases Hailuo 02, a video generation model featuring a new architecture for improved physical accuracy and native 1080p resolution.
- May 26 MiniMax Hailuo 2.3 Enhances Complex Visual Performance and Introduces Media Agent MiniMax releases version 2.3 of its video generation model, featuring improved handling of complex physical interactions and a new agent for automated media editing.
- May 26 MiniMax Debuts M2 Model for Enhanced Video and Agentic Performance MiniMax releases its M2 model, focusing on improved reasoning capabilities and higher-quality video generation for complex creative tasks.
About MiniMax Hailuo
MiniMax's video work used to be sold as Hailuo, a per-clip model family (Hailuo 2.3 and 2.3 Fast are still priced per video on the API). In 2026 the lab moved to H3, an omni-modal model that takes text, images, video and audio references and returns video. The API prices H3 per second of output: $0.08/s at 768P and $0.13/s at 2K, with H3-Max at $0.05/s for 480P. H3-Regeneration takes a 768P render and regenerates it at 2K for $0.05/s, which makes a practical workflow of generating cheap variants at 768P and upscaling only the keepers.
For filmmakers the useful parts are image-to-video with a first or last frame, reference generation (up to 9 images, 3 video clips and 3 audio clips per request for character, motion, camera, style or voice), and video-to-video motion transfer. Weak spots are the 15 second cap, integer-second durations only, no lip-sync tool, and an open-weight license that is more restrictive than standard open source. It fits B-roll, previs, retry-heavy shot generation and pipelines that need an API. It is not a finished editing or agent suite.
What's new in MiniMax Hailuo
-
H3 weights published on Hugging Face
MiniMaxAI/MiniMax-H3 released under the MiniMax H3 Community License; local output is 768P, 2K stays API-only.
Source: huggingface.co ↗ -
H3 offered inside Luma
Luma's pricing page lists MiniMax H3 and H3 Max among the third-party models available in its plans.
Source: lumalabs.ai ↗ -
Paid music APIs closed to new users
Music-3.0 and earlier paid music endpoints stop accepting new users from 20 August 2026, per the pricing page.
Source: platform.minimax.io ↗ -
H3 video model launched
Omni-modal video model: 4 to 15 second clips, 2K default, native stereo audio, text, image, video and reference modes.
Source: minimax.io ↗
Models available in MiniMax Hailuo
-
Hailuo 2.3 Latest — MiniMax Previous generation, priced per video: 768P 6s $0.28, 1080P 6s $0.49 Also: Hailuo-2.3-Fast
-
-
-
-
$ · $$ · $$$ = relative cost against other models of the same type, on fal.ai list prices. Cost inside MiniMax Hailuo depends on its own plans and credits.
Key Features
- H3 text-to-video, image-to-video and reference generation at 4 to 15 seconds (integer seconds only)
- First or last frame control with 0 to 2 tagged images in image-to-video
- Reference generation with up to 9 images, 3 video clips and 3 audio clips per request
- Video-to-video motion transfer and controllable editing
- 768P or 2K output; H3-Regeneration upgrades a 768P render to 2K
- Prompt camera tokens [pan], [zoom] and [static]
- Open weights on Hugging Face for local 768P inference
- Official MiniMax-MCP server with an async generate_video tool
- Speech 2.8 TTS and voice cloning on the same API
When to reach for it — and when to skip
Reach for it when…
- Published per-second pricing that is easy to budget ($0.08/s at 768P, $0.13/s at 2K)
- Cheap draft-then-upscale path: 768P variants, then $0.05/s regeneration to 2K for the keepers
- Open weights mean you can test locally before paying for hosted 2K
Skip it when…
- Hard 15 second cap per clip, and durations are whole seconds only
- Local open weights top out at 768P; 2K needs the hosted API
- The open-weight license is restrictive
- Model naming is confusing (H3, H3-Max, H3-Regeneration, Hailuo 2.3), which makes budgeting and citing versions harder
Best For
✓ Ideal for
- Retry-heavy B-roll: generate several 768P variants, then regenerate the best one at 2K
- Image-to-video shots that need a locked first frame from a still or storyboard panel
- Keeping one character, product or voice consistent across shots with reference inputs
- Pipelines that need an API and predictable per-second costs
✗ Not built for
- Dialogue scenes that need lip-sync or a talking-head tool
- Teams that need a licence they can read in one sitting: check the H3 Community License first
Working Tips from Filmmakers Using MiniMax Hailuo
- 01 Budget at 768P first. At $0.08/s a 10 second draft is $0.80; only regenerate winners at $0.05/s for 2K.
- 02 Lock the composition with a first-frame image. The first frame anchors the shot far better than prompt text alone.
- 03 Test one clip for audio before promising sound to a client.
MiniMax Hailuo pricing in 2026
| Plan | Price | What you get |
|---|---|---|
| H3 768P (API) | $0.08 per second of output |
|
| H3 2K (API) | $0.13 per second of output |
|
| H3-Max 480P (API) | $0.05 per second of output |
|
| H3-Regeneration (API) | $0.05 per second of regenerated output |
|
| Hailuo-2.3 768P 6s (API, legacy) | $0.28 per video |
|
Prices checked against MiniMax Hailuo's pricing page on 9 Oct 2026. Vendors change plans often, so confirm before you buy.
MiniMax Hailuo vs Kling AI vs Luma Dream Machine vs Pika
| MiniMax Hailuo | Kling AI | Luma Dream Machine | Pika | |
|---|---|---|---|---|
| Price | API: H3 from $0.08/s (768P) or $0.13/s (2K); see the vendor's pricing page for consumer plans. | See the vendor's pricing page | From $30/month (Plus), or $25/month billed yearly ($300/yr) | Starter $10/mo ($8/mo billed yearly); Creator $35/mo; Enterprise custom |
| Infinite canvas | Not confirmed | Partial | Yes | No |
| Node / flow builder | No | Not confirmed | Partial | No |
| AI agent | Yes | Not confirmed | Yes | Partial |
| Chat assistant | Not confirmed | Not confirmed | Yes | Partial |
| Team collaboration | Not confirmed | Not confirmed | Yes | Not confirmed |
| Text to image | Yes | Yes | Yes | Yes |
| Image editing | Not confirmed | Not confirmed | Yes | Yes |
| Text to video | Yes | Yes | Yes | Yes |
| Image to video | Yes | Yes | Yes | Yes |
| Video to video | Yes | Yes | Yes | Yes |
| Native audio in video | Yes | Yes | Not confirmed | Yes |
| Lip-sync | Not confirmed | Yes | Yes | Partial |
| Voice / TTS | Yes | Partial | Yes | Yes |
| Music & SFX | Partial | Not confirmed | Yes | Yes |
| Upscaling | Partial | Not confirmed | Yes | Not confirmed |
| Character consistency | Yes | Yes | Yes | Yes |
| Camera control | Partial | Yes | Yes | Partial |
| Custom training | Not confirmed | Not confirmed | Partial | Not confirmed |
| Third-party models | No | No | Yes | Yes |
| Public API | Yes | Yes | Yes | Yes |
| MCP server | Yes | Not confirmed | Partial | Yes |
| Mobile app | Not confirmed | Not confirmed | Not confirmed | Yes |
| Desktop / local | Yes | No | No | No |
| Open source | Partial | No | No | No |
| Free tier | Partial | Yes | Not confirmed | Partial |
| Commercial use | Not confirmed | Yes | Yes | Partial |
Each column comes from that tool's own researched page. "Not confirmed" means the vendor doesn't say.
Open the full MiniMax Hailuo vs Kling AI vs Luma Dream Machine vs Pika comparison →Use Cases
Tags
Creators Using MiniMax Hailuo
AI Filmmaker - Cinematic Storyteller - Google Flow - Seedance - Pixverse - Kling - Sora - Runway - Hailuo - NanoBanana - GPT - Flux - Gemini
AI FILMAKER
AI Filmmaker • Video Editor • AI Content Creator
Sources
Facts on this page were checked against MiniMax Hailuo's own pricing, product and changelog pages on , then cross-checked with independent coverage where vendor pages were silent. Anything we couldn't confirm is marked "Not confirmed" rather than guessed.
- MiniMax API pricing overview (redirect from minimax.io/pricing) platform.minimax.io
- MiniMax video generation guide platform.minimax.io
- MiniMaxAI/MiniMax-H3 model card huggingface.co
- MiniMax news (H3 listed as new) minimax.io
- MiniMax agent product page minimax.io
- MiniMax Design product page minimax.io
- OpenSourceForU: MiniMax releases H3 opensourceforu.com
- Luma pricing (lists MiniMax H3 models) lumalabs.ai
Spotted something out of date? Tell us and we'll re-check it.
Discussion
No comments yet — be the first.