MiniMax H3
MiniMax H3, also called Hailuo 3.0, is the Shanghai lab's omni-modal video model from 31 July 2026: one transformer that takes text, images, video and audio in and returns video with native stereo sound, up to 2K and 15 seconds, in seven shapes. MiniMax open-sourced the base weights three days later. On fal it costs $0.05 a second at 480p, $0.06 at 768p, $0.13 at 2K and $0.16 at 4K, which MiniMax pitches as a fraction of mainstream models' rates.
Anvisha Pai, Co-founder & CEO, Voyager
MiniMax · released
›What we asked for
A self-portrait of MiniMax H3, the open-weight video model with native sound: a jazz trio playing in a smoky club, close on the drummer's brushes then the double bass, warm light, the music itself audible and in time.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
At a glance
- Price
- $0.06 a second at 768p, $0.13 at 2K. Our five-second 768p clips cost $0.30.
- Sound
- Native stereo audio generated with the picture.
- Length
- 5 to 15 seconds.
- Sizes
- 480p, 768p, 2K or 4K in seven shapes including 21:9.
- Open weights
- The base model is on Hugging Face under MiniMax's community licence; the prompt preprocessor and 2K upscaler stay API-only.
Other Hailuo models: H3 Max Turbo, H3 Max, Hailuo 2.3. Compare all 4.
Anvisha's take
MiniMax H3 is the cheap model with sound that mostly holds up. The title card is crisp, the harbour and market are convincing, the physics clip is tidy, and the audio track is there, with the woman's line delivered on the beat. The pour-over came back as a close-up of a cup and a spoon with the kettle out of frame, and its self-portrait split into a drummer and a cellist in one five-second clip. $0.30 a clip at 768p, one to three minutes.
MiniMax H3 real output: the same 9 clips every model gets
Every video model on this site gets the same 9 clips, so you can compare like with like. These are the clips MiniMax H3 returned, one try each, nothing re-rolled or retouched. Each tile shows one frame of the clip until you press play. Generated on . The 9 clips took 52 to 182 seconds each and cost $2.70 in total.
A clean product shot that holds its shape through a full slow rotation.
›What we asked for
A matte black ceramic pour-over coffee dripper rotating slowly on a white turntable, soft studio light, seamless white background, product video, no text.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
A face, lip sync and a spoken line, where the model makes audio.
›What we asked for
A woman in her 40s with short grey hair stands in a bright kitchen, looks at the camera and says 'Good morning, the coffee is ready', natural window light, handheld feel, she smiles at the end.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
Set text that stays spelled correctly while it moves.
›What we asked for
A title card reading 'MODA SUMMER SESSIONS' animates in over slow-motion footage of a festival crowd at dusk, then a smaller line 'Friday 12 June' fades in beneath it. Bold condensed type, tangerine on navy.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
A steady drone push-in with stable geometry.
›What we asked for
Drone shot pushing slowly forward over a coastal town at sunrise, terracotta roofs, fishing boats in the harbour, long shadows, cinematic and steady.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
A spill in slow motion: does water behave like water.
›What we asked for
A glass of water tips over on a wooden table and spills in slow motion, water and three ice cubes sliding across the wood, macro lens, morning light.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
Hands, a kettle and a steady pour without extra fingers or teleporting objects.
›What we asked for
Close-up of hands pouring hot water from a gooseneck kettle in slow circles into a pour-over coffee dripper, steam rising, shallow depth of field, warm cafe light.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
Many people and objects moving at once without merging.
›What we asked for
A crowded Saturday farmers market seen from slightly above, shoppers with tote bags, a dog on a leash, a busker with a guitar, bunting swaying, late morning light, gentle camera drift.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
A named style held across every frame rather than drifting to photoreal.
›What we asked for
A lighthouse on a rocky cliff in a storm, in the style of a Japanese ukiyo-e woodblock print, waves crashing in flat colour with visible wood grain, animated like moving paper cutouts.
minimax/h3/text-to-video
aspect_ratio: 16:9, resolution: 768P, duration: 5
returned 1344 by 768 pixels
Bring a still product photo to life without changing the product.
›What we asked for
The coffee dripper slowly fills as coffee drips through it, steam rising, the camera drifting gently around it, the dripper itself unchanged. (starting from The Nano Banana Pro product shot from the image fixture set.)
minimax/h3/image-to-video
resolution: 768P, duration: 5
returned 768 by 768 pixels
How the cost under each picture was worked out: fal's MiniMax H3 rate of $0.06 per second at 768p (fal model page, read 2026-09-09) for the 5 seconds requested; fal does not return the charge.
Compare MiniMax H3
MiniMax H3 pricing
| Where | Unit | Price | Source |
|---|---|---|---|
| fal | Second at 768p $0.05 at 480p, $0.13 at 2K, $0.16 at 4K; the reference endpoint adds $0.08 per image beyond five. | $0.06 | fal model page, read |
| fal | Second at 2K | $0.13 | fal model page, read |
| fal | Five-second 768p clip | $0.30 | fal model page, read |
Is MiniMax H3 free?
The Hailuo app gives daily free credits, with more during launches, and MiniMax H3's base weights are free to download under the MiniMax H3 Community License. The API and hosts charge per clip or per second.
- •Hailuo app and web: daily free credits.
- •H3 open weights: the 33-billion-parameter base model on Hugging Face and ModelScope; the prompt preprocessor and 2K upscaler stay API-only.
- •fal and MiniMax's platform: per clip or per second.
MiniMax H3 announcement, read
Where to use MiniMax H3
- Hailuo app · MiniMax's own app, daily free credits.
- Open weights · The H3 base model under the MiniMax H3 Community License.
- MiniMax platform · MiniMax's API.
- fal · Per-second by resolution; text, image and reference endpoints.
- Voyager · Runs inside Voyager, the agent for creative work. Private preview by waitlist.
Voyager
Run MiniMax H3 inside Voyager
Voyager is an agent for creative work that runs every model in this reference, MiniMax H3 included. Brief it, and it picks the model, makes the thing and hands back something you can edit. Create an account to get started.
What we noticed running MiniMax H3
- •Reinterpreted the pour-over as a hand tilting a cup of black coffee beside an espresso, kettle and filter gone. See the hands at work clip.
- •Self-portrait: The self-portrait cut between a drummer's brushes and a cellist's hands inside one clip; H3 edits when it wants to.
- •The glass tips gently and the ice slides out with no splash; restrained rather than wrong. See the physics clip.
- •Every clip landed at 1344 by 768, a slightly wider frame than the 1280 by 720 most vendors return for 16:9.
MiniMax H3 limits and API parameters
| Limit | Value |
|---|---|
| Length | 5 to 15 seconds. fal API reference, read |
| Resolution | 480p, 768p, 2K or 4K; the pixel size at a given label is not fixed. fal API reference, read |
›Every setting, for developers
fal (minimax/h3/text-to-video)
fal API reference, read
| Parameter | Type | Values | Default |
|---|---|---|---|
| prompt | string | Any text Required. | |
| duration | integer | 5 to 15 seconds | 5 |
| resolution | enum | 480P, 768P, 2K, 4K | 2K |
| aspect_ratio | enum | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | 16:9 |
| prompt_expansion_mode | enum | fast, balanced, quality | balanced |
| seed | integer | Any integer | |
| image_url | URL | Starting frame Image-to-video endpoint. |
What MiniMax H3 will not make
MiniMax's models run behind fal's safety checker and fal's partner terms for commercial use; MiniMax's own platform terms bar sexual content, minors, violence, hate and deceptive depictions of real people, and the Hailuo app labels outputs as AI-generated. H3's open weights carry the MiniMax H3 Community License.
fal API reference, read
Alternatives to MiniMax H3
Frequently asked questions
What is MiniMax H3?
MiniMax H3, or Hailuo 3.0, is MiniMax's omni-modal video model from 31 July 2026, returning video with native stereo sound at up to 2K and 15 seconds, with open base weights and per-second pricing from $0.05 on fal.
How much does MiniMax H3 cost?
$0.05 a second at 480p, $0.06 at 768p, $0.13 at 2K and $0.16 at 4K on fal, read on 9 September 2026.
Is MiniMax H3 open source?
The base weights are, under the MiniMax H3 Community License since 3 August 2026. The prompt preprocessor and the 2K upscaler that round out the API version are not.
How this page is made
Prices, limits and settings are read from the linked vendor and host pages on the dates shown. Every picture was generated by us through fal with the request shown under it, and the original files are kept unedited. A generation that failed is shown as a failure. Model and vendor names are used to identify the products; no vendor imagery appears on this page.
