Eight versions of MiniMax's video model, from 480P drafts in seconds to native 2K, each with native audio and up to 15 seconds from text, frames or references.
MiniMax H3 is a video generation model from MiniMax that turns a text prompt, a first or last frame, or image, video and audio references into clips of up to 15 seconds with native audio. Segmind runs eight versions through one API key: H3, H3 Max, H3 Highspeed, H3 Lite, H3 Max Turbo, H3 Max Multi-Angle, H3 Optimised and Creatify's Boreal-H3, from 480P drafts to native 2K.
Same model family, different trade-offs. Every version generates sound with the picture and runs through the same Segmind API key. Open any model page to run it in the playground or copy its API request.
MiniMax · Hero shots
The full model at native 2K, with strong instruction following and accurate on-screen text.
Native 2K from a text prompt, surf and wind generated with the picture.
MiniMax · The hardest prompts
The top quality tier, tuned for crowded scenes, fast action, fluids, smoke and reflections, with prompt expansion modes.
Liquid nitrogen fog rolling off a bench, 6 s at 768P.
MiniMax · Fast iteration
Balances quality and speed at 480P or 768P. MiniMax quotes about 11 to 13 seconds to generate a 5-second clip.
A 10-second domino chain, the camera tracking low beside it.
MiniMax · Drafts at volume
Tuned for speed and cost at 480P or 768P. MiniMax quotes about 8 to 9 seconds to generate a 5-second clip.
A silk gown on the runway, 6 s at 768P.
MiniMax · Exploring many shots
The speed-tuned tier of H3 Max, with first and last frame keyframing and legible on-screen titles.
Kinetic type, spelled right, from one prompt.
MiniMax · Camera moves on a still
Freezes the scene in one photo and moves a virtual camera through it along 2 to 12 timed keyframes of orbit, elevation and distance.
One photo of a welder; the sparks stay frozen while the camera cranes overhead.
MiniMax · Faster turnaround
MiniMax H3 on an optimised inference stack. Renders 768p natively and upscales to 1080p or 2k, with a high or low quality switch.
A line of Japanese dialogue, voice and room sound in the same pass.
Creatify · Ads and social
Creatify's post-train of the H3 base, with a handheld, phone-footage realism instead of a graded cinematic look.
Vertical, handheld, and she talks to the camera.
Each MiniMax H3 version has a model page per mode. The examples below are real requests and outputs, prompt included.
01
Describe the subject, the action, the camera and the sound. Every version takes a plain prompt, in any of six aspect ratios from 21:9 to vertical 9:16.
Prompt
Close-up on a cast-iron skillet over a gas flame: a knob of butter slides and foams, then a thick steak is laid in and sears hard, fat spitting and ribbons of smoke curling up. The camera holds steady on the crust forming.
9:16, 4 s, 768P on MiniMax H3 Highspeed
02
Give a first frame, a last frame or both, and the model animates the motion between them. Useful for product reveals and for cutting generated shots together.
Prompt
The folded paper crane rises and straightens on the walnut table, its creased wings spreading outward and lifting until it stands upright, sharp folds catching the light.
16:9, 5 s, 768P on MiniMax H3 Max Turbo
03
Send up to 9 images, 3 videos and 3 audio clips in one request. Here, two product photos of one toy keep its shape and colours fixed while the prompt sets the scene; a clip can supply a camera move and an audio file a voice.
Prompt
A red wind-up tin robot with a key in its back marches across a wooden desk, arms swinging, then stops and tilts its head as a paper clip topples beside it.
Adaptive ratio, 6 s, 768P on MiniMax H3 Max
Every version bills per second of output at the resolution you choose. Current rates are on each model page's pricing tab.
| Version | Best for | Resolution | Length | Inputs | References |
|---|---|---|---|---|---|
| MiniMax H3MiniMax | Hero shots | 768P, 2K | 4 to 15 s | Text, frames, references | 9 images, 3 videos, 3 audio |
| MiniMax H3 MaxMiniMax | The hardest prompts | 480P, 768P | 5 to 15 s | Text, frames, references | 9 images, 3 videos, 3 audio |
| MiniMax H3 HighspeedMiniMax | Fast iteration | 480P, 768P | 4 to 15 s | Text, frames, references | 9 images, 3 videos, 3 audio |
| MiniMax H3 LiteMiniMax | Drafts at volume | 480P, 768P | 4 to 15 s | Text, frames, references | 9 images, 3 videos, 3 audio |
| MiniMax H3 Max TurboMiniMax | Exploring many shots | 480P, 768P | 5 to 15 s | Text, first and last frame | None |
| MiniMax H3 Max Multi-AngleMiniMax | Camera moves on a still | 480P, 768P, 1080P | 5 to 15 s | One image plus a camera path | None |
| MiniMax H3 OptimisedMiniMax | Faster turnaround | 480p to 2k | 5 to 15 s | Text, frames, references | 4 images, 1 video |
| Boreal-H3Creatify | Ads and social | 768p, 1088p, 2k | 1 to 15 s | Text, frames, references | 9 images, 3 videos, 3 audio |
Every version outputs 24 fps MP4 with audio, in 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16 (H3 Max Multi-Angle follows the input photo). See Segmind pricing for plans and credits.
MiniMax H3 is MiniMax's third-generation video model, also known as Hailuo 3. It reads text, images, video and audio as one context and returns clips of up to 15 seconds with dialogue, effects and ambience generated in the same pass. Segmind serves every MiniMax H3 version through one pay-as-you-go API.
Use MiniMax H3 for native 2K hero shots, H3 Max for the hardest prompts, H3 Highspeed for fast iteration and H3 Lite for drafts at volume. H3 Max Turbo explores many shots quickly, H3 Max Multi-Angle moves a camera around one photo, and Boreal-H3 gives a handheld, phone-footage look for ads.
MiniMax H3 Highspeed balances quality and speed, and MiniMax H3 Lite is tuned for speed and cost. Both render at 480P or 768P. MiniMax quotes about 11 to 13 seconds for a 5-second Highspeed clip and 8 to 9 seconds on Lite. Both take text, frames and references.
MiniMax H3 Optimised serves MiniMax's H3 model on an optimised inference stack for faster turnaround. It renders 768p natively and upscales to 1080p or 2k, takes up to 4 reference images and one reference video, and has a high or low quality switch. Clips run 5 to 15 seconds with native audio.
Boreal-H3 is Creatify's post-train of the MiniMax H3 base, narrowed for advertising and short-form social video. It favours a handheld, phone-footage realism over a graded cinematic look, renders 1 to 15 seconds at 768p, 1088p or 2k, and takes the same image, video and audio references as MiniMax H3.
MiniMax H3 Max Multi-Angle freezes the scene in one photo and moves a virtual camera through it. You give 2 to 12 timed keyframes of orbit, elevation and distance, and every person, object and spark holds still while the camera travels, at 480P, 768P or 1080P for 5 to 15 seconds.
A MiniMax H3 video can be up to 15 seconds from one request. MiniMax H3, H3 Highspeed and H3 Lite start at 4 seconds; H3 Max, H3 Max Turbo, H3 Max Multi-Angle and H3 Optimised start at 5 seconds; Boreal-H3 accepts any whole number of seconds from 1 to 15.
Yes. The MiniMax H3 reference-to-video models take up to 9 images, 3 videos and 3 audio clips in one request, so a character, a camera move and a voice or soundtrack can each come from their own reference. Each type is limited to 15 seconds in total.
MiniMax H3 is billed per second of output at the resolution you choose, with no subscription. On the reference-to-video models, seconds of reference video and images beyond the free allowance are billed too. Each MiniMax H3 model page lists its current per-second rates on the pricing tab.
Create a Segmind API key, then POST your prompt to https://api.segmind.com/v1/ followed by the model slug, such as minimax-h3-text-to-video, with the key in the x-api-key header. For longer clips submit to the /v2 endpoint and poll the request. Every MiniMax H3 model page has curl, Python and JavaScript examples.
One API key runs every MiniMax H3 version, so a shot can move from a quick draft to a final render without changing your code beyond the model slug.
Last updated