MiniMax H3
MiniMax H3-specific models, limits, and pricing for the shared AI Video generation endpoint.
MiniMax H3 runs on the shared video generation endpoint. For the endpoint, request flow, content items, polling, task listing, and error codes, see the AI Video overview. This page covers only what is specific to MiniMax H3.
Models
| Model | Best for | Rate limit |
|---|---|---|
MiniMax-H3 | Clips with a native sound track, closing-frame and audio reference | 5 RPM |
Pass the model in the model field of POST /v1/video-generation/generate. The literal is case-sensitive: MiniMax-H3.
Limits
| Limit | MiniMax H3 |
|---|---|
| Resolution | 768p, 2k |
| Duration | 4-15s, whole seconds only |
| Aspect ratios | 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 |
| Prompt length | up to 2500 characters, required |
| Reference images | up to 5 per request |
| Reference videos | up to 3 per request |
| Reference audio | up to 3 per request |
inputVideoDuration | 2-15s, required with a reference video |
768p sizes the short edge to 768 pixels — 1344×768 at 16:9. 2k sizes the long edge to 2560 pixels. Both are exclusive to MiniMax H3: every other model on the shared endpoint rejects them, and MiniMax H3 in turn rejects 480p, 720p, and 1080p.
MiniMax H3 accepts first_frame, last_frame, or both. It is the only model on this endpoint that takes a last_frame on its own — everywhere else a closing frame is only valid alongside an opening one, so a "land on this shot" request needs no filler first frame here. Frame roles and reference roles (reference_image, reference_video, reference_audio) cannot be combined in one request.
In image-to-video the provider derives the aspect ratio from the input image, so ratio is accepted and then ignored. Set the framing by cropping the image you send rather than by the parameter.
MiniMax H3 takes 768p and 2k only — 480p, 720p, and 1080p return
400, and no other model accepts 768p or 2k. Duration is 4-15s. The
4:5 and 5:4 aspect ratios are rejected, and ratio is ignored entirely
for image-to-video. seed and generateAudio are accepted for
compatibility but have no effect: generation is not seed-reproducible, and
the output always carries an audio track. reference_image is capped at 5
per request, not the 9 the shared endpoint allows elsewhere.
Pricing
Credits scale with resolution and duration. The total is rounded up to a whole credit.
| Resolution | Credits / second | 4s | 10s | 15s |
|---|---|---|---|---|
768p | 9.3075 | 38 | 94 | 140 |
2k | 15.1247 | 61 | 152 | 227 |
Call POST /v1/video-generation/estimate-credits with the MiniMax H3 model, resolution, and duration for the exact cost before generating. Credits are charged on task creation and refunded automatically on failure.
Example
curl -X POST "https://api.marswave.ai/openapi/v1/video-generation/generate" \
-H "Authorization: Bearer $LISTENHUB_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "MiniMax-H3",
"content": [
{ "type": "text", "text": "The camera pushes in slowly as the city lights come up" },
{ "type": "image_url", "role": "first_frame", "image_url": { "url": "https://example.com/open.jpg" } },
{ "type": "image_url", "role": "last_frame", "image_url": { "url": "https://example.com/close.jpg" } }
],
"resolution": "768p",
"ratio": "16:9",
"duration": 8
}'See the AI Video overview for the full parameter list and the text-to-video / image-to-video request shapes.