MiniMax H3 Highspeed is available on EmpirioLabs. It is MiniMax's fast H3 model: clips of 4 to 15 seconds at 768p with a native stereo audio track, from a prompt, an opening frame, a first and last frame, or a mix of image, video, and audio references. A 4 second clip from one image finished in about half a minute end to end in our tests.
How it compares
| MiniMax H3 Highspeed | MiniMax H3 Lite | MiniMax H3 | |
|---|---|---|---|
| Resolutions | 768p | 480p, 768p | 768p, 2K |
| Clip length | 4 to 15 seconds | 4 to 15 seconds | 4 to 15 seconds |
| Native stereo audio | Yes | Yes | Yes |
| Prompt expansion and 2K regeneration | No | No | Yes |
How to call MiniMax H3 Highspeed
Submit the job, then poll it. This animates one image as the opening frame:
curl -X POST https://api.empiriolabs.ai/v1/videos/generations \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-h3-highspeed",
"prompt": "The front wheel slowly turns as a breeze moves the leaves overhead.",
"image": "https://example.com/bicycle.jpg",
"duration": 4
}'
# {"job_id": "...", "status": "processing", "poll_url": "/v1/jobs/..."}
curl https://api.empiriolabs.ai/v1/jobs/JOB_ID \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY"The mode follows what you attach: no image is text to video, one image is the opening frame, an image plus a last_frame is first and last frame, and two or more images, or any video or audio reference, is reference to video.
Operational notes
durationtakes whole seconds from 4 to 15. A value outside that range is refused rather than clamped.- With
aspect_ratioleft atadaptive, an image-driven clip follows the image: a square photo gives a square video. Text to video uses 16:9 unless you choose a ratio. - Reference audio needs an accompanying reference image or video in the same request.
- Prompt expansion, 2K regeneration and the watermark belong to MiniMax H3. Requests that ask Highspeed for them are refused, so remove those fields when switching models.
Pricing
MiniMax H3 Highspeed is pay as you go, billed per second of the generated clip. A reference video is billed for its own duration in addition to the generated clip. Reference audio is not billed, and the first five reference images are included. Current rates are on the model page and the pricing page.
Try it
Run it in the Playground with no code, or read the API reference for every parameter. For 480p output, use MiniMax H3 Lite; for 2K, MiniMax H3.
