minimax/h3
MiniMax H3, served direct from MiniMax.
Text-to-video, image-to-video (a first frame, optionally a last frame) and reference-to-video (reference pictures, clips and sounds) in one model, every clip with sound; 4 to 15 seconds at 768P or 2K. Served through the async job API only: POST /v2/jobs, then poll.
Billed per second of output by resolution, from what the generator reports it rendered: 768P $0.096, 2K $0.156.
MiniMax H3 on this gateway: minimax/h3 (this page) · minimax/h3-highspeed · minimax/h3-lite · minimax/h3-max · minimax/h3-regeneration.
| Slug | minimax/h3 |
| Kind | video |
| Vendor | minimax |
| Endpoint | POST /v2/jobs |
Charged per job from what the generator reports it rendered — your Logs show the exact figure for each one; see balance and billing.
Calling it
curl https://api.sociaro.com/v2/jobs \
-H "Authorization: Bearer $SOCIARO_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: my-request-0001" \
-d '{
"model": "minimax/h3",
"input": {
"content": [
{"type": "text", "text": "The paper boat drifts down the rain-soaked street."},
{"type": "image_url", "image_url": {"url": "https://your-cdn.example.com/first.png"}, "role": "first_frame"}
],
"resolution": "768P",
"duration": 10
}
}'
# then poll the job id it returns until status is succeeded:
curl https://api.sociaro.com/v2/jobs/JOB_ID \
-H "Authorization: Bearer $SOCIARO_API_KEY"Parameters
MiniMax H3 — text, image and reference to video, with sound
| Field | Type | Required | Default | Values | What it does |
|---|---|---|---|---|---|
content | object[] | yes | — | text · image_url · video_url · audio_url | Typed parts: exactly one `{"type": "text", "text": …}` (up to 7000 characters), plus `image_url` parts with `role` first_frame, last_frame or reference_image, `video_url` parts with role reference_video and `audio_url` parts with role reference_audio. Each media part is `{"url": …}`, a public https URL or a `data:` URI. Up to 9 reference pictures, 3 clips and 3 sounds (each 2–15 s, 15 s in total per kind). Frames and reference parts cannot be mixed. |
resolution | string | yes | — | 768P · 2K | Sets the per-second rate. Any other value is a `400` before anything is rendered. |
duration | integer | yes | — | 4–15 | Whole seconds of output. |
ratio | string | no | adaptive | adaptive · 21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16 | A text-only request needs a concrete ratio (not `adaptive`). With a first or last frame the ratio follows the picture. |
Anything not listed here is refused rather than ignored, so a typo fails loudly instead of quietly producing something else.
Worked example
curl https://api.sociaro.com/v2/jobs \
-H "Authorization: Bearer $SOCIARO_API_KEY" -H "Content-Type: application/json" \
-H "Idempotency-Key: my-request-0001" \
-d '{ "model": "minimax/h3",
"input": {
"content": [ { "type": "text", "text": "A boy plays basketball by the sea at sunset; waves and a bouncing ball." } ],
"resolution": "2K", "duration": 8, "ratio": "16:9" } }'
curl https://api.sociaro.com/v2/jobs/JOB_ID -H "Authorization: Bearer $SOCIARO_API_KEY"Submit and poll. Served through the Jobs API only: POST /v2/jobs with the model in the envelope and
its fields in input, then poll GET /v2/jobs/{id} until status is succeeded. The finished MP4 is
outputs[0].url, a stable link on our CDN kept for ten years.
What is checked here. What the body decides is checked at submit, each a 400 invalid_input and no job: a field
the model's input_schema does not name, a value outside its list or range (a resolution this model does not offer),
a content part of the wrong shape, more parts of a kind than the generator takes (one first frame, one last frame, nine
reference pictures, three clips, three sounds), a first or last frame mixed with reference parts, a missing text part, a text-only request without a concrete
ratio, a callback_url inside input (the job's own callback_url goes at the top level, beside model and
input) and an mm_file:// url. The generator judges the media itself (the length of a clip or sound, a picture's
size and format) and refuses what it does not take before anything is billed.
prompt_enhance is not offered on these models.
How it is billed. Once, when the job completes, from what the generator reports it rendered: seconds of output at the rate of the resolution it rendered, plus any reference clips by their length at that resolution's input rate, plus each picture (first frame, last frame, reference) past the first five. Reference sounds are free. A failed job is not billed.
| 768P | 2K | |
|---|---|---|
| per second of output | $0.096 | $0.156 |
| per second of reference clip | $0.096 | $0.156 |
| per picture past the first five | $0.048 | $0.048 |
A 10 second 2K clip with 6 seconds of reference clips and 7 pictures: 10 × $0.156 + 6 × $0.156 + 2 × $0.048 = $2.592.