MiniMax H3 — 2K video with sound from text, frames or references
MiniMax H3 is MiniMax's multimodal video model: send a prompt, a first and last frame, or your own images, videos and audio, and get a 4–15 second clip with its own sound at 768P or 2K, from $0.073125 per second.
- Text to video
- First and last frame
- Reference to video
- Generated audio
- Up to 2K
- 4–15 seconds
Try MiniMax H3 before you write any code
Run MiniMax H3 with the exact request your code will send. Every parameter the API takes is here.
minimax-h3from $0.073125/s
Pick a mode, describe the shot, add frames or references if the mode needs them, and submit once. The MiniMax H3 clip arrives with the exact amount it cost.
The shot and its sound: subject, place, camera, light, dialogue and effects. Up to 7,000 characters.
Start from text alone, from a first and/or last frame, or from your own images, videos and audio. Frames and references cannot be combined.
768P costs less; 2K is the sharpest.
Required for text alone, and not adaptive there. With references, adaptive follows your media. With frames, the image sets the shape.
4–15 seconds.
On checks the prompt, images and reference-video frames before generating; flagged requests fail and are not charged.
Greyed-out options are available through the API only, not in the playground.
Estimated at current rates from the resolution and length. Reference video seconds and images beyond the fifth are added when you attach them.
Try one of these prompts
Generated with MiniMax H3 on SeedRouter, with its audio
MiniMax H3 pricing
MiniMax H3 is billed per second of video at a rate set by the resolution. Every rate below is the one that bills your account.
- MiniMax H3
- $0.365625/clip
- MiniMax H3
- 273seconds
All MiniMax H3 prices come from the rates that bill your account. Failed requests are not charged.
| Output | Billed quantity | Rate | Billing |
|---|---|---|---|
| Video · 768P | Output seconds plus the length of every reference video | $0.073125 / s | Fixed when the task is accepted |
| Video · 2K | Output seconds plus the length of every reference video | $0.118625 / s | Fixed when the task is accepted |
| Input images | Each image after the first 5; frames and reference images both count | $0.03575 | Fixed when the task is accepted |
| Failed requests | Any task that ends failed, including content-filter refusals | Free | Nothing is charged |
What is MiniMax H3?
MiniMax H3 is MiniMax's general-purpose video model, also called Hailuo 03. One MiniMax H3 call makes the picture and the sound together: spoken lines, ambience and effects that match the scene you describe.
You can drive MiniMax H3 three ways: a text prompt, a first and last frame, or up to 9 reference images, 3 videos and 3 audio clips the shot must follow. On SeedRouter you send MiniMax's own request format, get a task ID, and collect the MP4 when it is ready.
In our test, a 4-second 768P MiniMax H3 clip came back as MP4 (H.264) at 1344 × 768 and 24 fps, with a stereo AAC track at 32 kHz.
Two ways to use MiniMax H3
Call the API from your backend, or hand the job to a coding agent.
Call the API
Send the MiniMax H3 request from your server, keep the task ID, and read the finished clip from the task.
- 1Create an API key
- 2POST the request with your content
- 3Poll the task until the video is ready
Let an agent do it
Copy a ready prompt into Claude Code, Codex or Cursor. It asks before spending and downloads the result.
- 1Copy the prompt
- 2Paste it into your agent and describe the shot
- 3Approve the cost, then get the video
What MiniMax H3 is good at
MiniMax H3 is at its best when the shot, the sound and your own material have to line up.
Sound in the same render
Every MiniMax H3 clip has a stereo track: dialogue, ambience and effects. No voiceover pass, no syncing.
2K when it matters
Pick MiniMax H3 at 2K for the final cut and 768P for drafts. Same request, one field changes.
Start and end where you want
Give MiniMax H3 a first frame, a last frame or both, and the clip moves between them.
Your own material
Mix up to 9 images, 3 videos and 3 audio clips. MiniMax H3 keeps the look, motion or voice you give it.
A voice you choose
Attach a reference audio clip and write the line; the speech follows that voice.
Up to 15 seconds
One MiniMax H3 call covers a whole short scene, from 4 to 15 seconds.
What teams use MiniMax H3 for
Product spots
Put a real product photo into a MiniMax H3 reference request and get a short spot with sound.
Talking scenes
A character speaks a written line in a referenced voice, with the room sound to match.
Social clips
MiniMax H3 clips in vertical 9:16 for Reels and TikTok or wide 16:9 for YouTube, audio already in.
Storyboard to motion
Turn a key frame and an end frame into a MiniMax H3 shot that joins them.
Restyle footage
Feed MiniMax H3 a reference video and describe a new look, setting or light for the same motion.
Trailers and teasers
Cut 2K MiniMax H3 shots into a teaser, each with its own sound bed.
Start with MiniMax H3 in four steps
- 01
Create a key
Sign in, top up any amount, and create an API key. Credits never expire.
- 02
Send one request
POST MiniMax's request body to /v1/videos/generations with model minimax-h3 and your content.
- 03
Poll the task
Keep the MiniMax H3 task ID and check it every 10–20 seconds until the status is completed.
- 04
Download the video
Save the hosted MiniMax H3 URL from the task to your own storage.
MiniMax H3 at a glance
The facts that decide whether MiniMax H3 fits the job.
| Feature | MiniMax H3 |
|---|---|
| Resolution | 768P, 2K |
| Clip length | 4–15 seconds |
| Inputs | Text, first and last frame, or images, videos and audio |
| Audio | Always generated: dialogue, ambience, effects |
| Request format | MiniMax's official body |
| Billing | Per second, by resolution |
Need a cheaper 480p draft? See SD Video. Need clips longer than 15 seconds? See Seedance 2.5.
What to know before you start
Frames or references, not both
A MiniMax H3 request uses either a first/last frame or reference media. Text alone needs an explicit aspect ratio.
Media is sent as URLs
Send MiniMax H3 images, clips and audio as public HTTP(S) links, not base64. The Playground uploads for you.
Results arrive by task
A MiniMax H3 request returns a task ID first. Poll it until the video is ready.
Content filter is on by default
MiniMax H3 prompts, images and reference-video frames are checked before generation. A flagged request fails and costs nothing.
Why run MiniMax H3 on SeedRouter
Pay per second of video
MiniMax H3 is billed on the seconds you ask for, at the rates in the table above. No subscription, no minimum.
Failed requests are free
If a MiniMax H3 task ends failed, nothing is charged. You only pay for videos that are delivered.
Credits never expire
Top up once and use the balance whenever you need it.
Hosted results
Finished MiniMax H3 clips come back as URLs on our CDN. Keep a copy of anything you need long term.
One balance for every model
The same top-up pays for video, image and text models in the catalogue.
Other models on SeedRouter
Same balance, same task-based workflow.
MiniMax H3 guides
Practical guides to pricing, prompting and comparing the model.
MiniMax H3 costs $0.08 a second at 768p and $0.13 at 2K on the official API. Its open weights are free, need about 108 GB and exclude the US, EU, UK and Korea.
How to write MiniMax H3 prompts: the official three-part structure, dialogue tags, shot timing and reference labels, tested on four clips.
MiniMax H3 vs Seedance 2.5: a blind ranking calls them tied. Seedance 2.5 leads on anatomy, lip sync and 30-second clips; H3 costs far less per minute.
MiniMax H3 FAQ
What is MiniMax H3?+
MiniMax H3, also called Hailuo 03, is MiniMax's multimodal video model. It makes one clip of 4 to 15 seconds at 768P or 2K, with dialogue, ambience and effects generated together with the picture. You can start from text, from a first and last frame, or from up to 9 images, 3 videos and 3 audio clips.
How much does MiniMax H3 cost?+
A 5-second 768P text-to-video clip costs $0.365625. MiniMax H3 is billed per second at a rate set by the resolution; reference videos add their own length, and each input image after the fifth costs $0.03575. The pricing table on this page shows every current rate.
Is MiniMax H3 free?+
Not on SeedRouter. MiniMax H3 is paid per second of video, and new accounts get a small credit to try it. There is no subscription: you top up any amount, credits never expire, and failed requests are not charged.
How do I use the MiniMax H3 API?+
Create a SeedRouter key and POST MiniMax's official body to /v1/videos/generations with model minimax-h3: a content list with your prompt and any media, plus resolution, duration and, for text alone, ratio. You get a task ID back; poll /v1/tasks/{id} until the video is ready.
Does MiniMax H3 generate audio?+
Yes. Every MiniMax H3 clip has a stereo audio track made from the same prompt: spoken lines, room tone and effects. Write the dialogue into the prompt, and attach a reference audio clip if the voice should follow it.
How do references work in MiniMax H3?+
Add reference images, videos or audio to the content list with the roles reference_image, reference_video and reference_audio, then describe the scene around them. References cannot be mixed with a first or last frame in the same request.
Does MiniMax H3 support 2K?+
Yes. MiniMax H3 renders at 768P or 2K; you pick one per request. 2K costs more per second, so many teams draft at 768P and render the final shot in 2K.
Is MiniMax H3 content filtered?+
On SeedRouter the content filter is on by default: the prompt, images and frames from reference videos are checked before MiniMax H3 runs. A flagged request fails without charge. Through the API you can turn the filter off with content_filter set to false.
Generate your first MiniMax H3 clip
Try MiniMax H3 in the Playground, then move the same request into your code.