MiniMax H3 API
MiniMax H3 API gives APINEED callers one asynchronous endpoint for text-to-video, first-frame animation, and start-to-end frame generation at 768P or 2K.
MiniMax documents H3 as a unified video-and-audio foundation model with text, image, video, and audio conditioning. The current APINEED browser workflow exposes only the production paths verified end to end: text, one reference image, or two frame images. Broader reference-video and reference-audio inputs remain documented upstream capabilities rather than controls this page pretends to support.
Playground
Generate with MiniMax H3, review the output, and use the same request shape in your application.
Request preview
{
"model": "minimax-h3",
"prompt": "A cinematic product film begins with a quiet tabletop at dawn, then the camera arcs around a glass sculpture while soft wind moves the surrounding fabric. Natural reflections, restrained motion, no text overlay.",
"duration": 5,
"size": "768P",
"metadata": {
"ratio": "16:9"
}
}Providers
MiniMax documents the broader model capability. APINEED separately defines the model ID, endpoint, parameter whitelist, authentication, and billing behavior shown on this page.
Discount
Compare the TokenDance channel anchor with APINEED per-second pricing. The selected duration is the request billing multiplier. [3][4]
| Billing unit | Official price | APINEED price | Savings |
|---|---|---|---|
| 1 generated second | $0.074627 | $0.0686578% off | 8% off |
Availability
APINEED monitors MiniMax H3 API availability. Applications should preserve request and task IDs and handle upstream failures explicitly.
Four-to-fifteen-second generation
The APINEED adapter validates whole-second duration from four through fifteen seconds before creating the asynchronous H3 task, preventing unsupported or unbounded billing inputs. [2][4]
768P and 2K output
Callers select 768P or 2K through the top-level size field. The browser does not invent additional quality levels or dimension strings that the current route has not verified. [2][4]
Text and frame-guided workflows
A request without images is text-to-video, one image becomes the first frame, and two images become the starting and ending frames used to guide the generated transition. [2][4]
Aspect ratio for text generation
Text-only requests include metadata.ratio using one of six documented compositions. Frame-guided requests omit ratio so the uploaded image geometry can remain authoritative. [2][4]
Asynchronous completion
APINEED returns a task identifier from POST /v1/videos. Applications retain that identifier and poll the shared task endpoint until a completed URL or a clear upstream error is available. [3][4]
Directing MiniMax H3 through APINEED
Write a timed shot. Describe the initial frame, action order, camera movement, lighting changes, and final composition within the selected duration.
Use two frames with intent. Choose an end frame that the subject and camera can plausibly reach from the first image in four to fifteen seconds.
Let frame geometry lead. When supplying a reference image, crop it to the desired composition because APINEED deliberately omits a competing aspect-ratio field.
Quick Start
Call MiniMax H3 with the verified APINEED endpoint and model-specific request shape.
Create an API key
Store your APINEED key server-side as API_NEED_API_KEY.
export API_NEED_API_KEY=sk-apineed-v1-...Send the request
Use minimax-h3 with POST /v1/videos.
curl https://apineed.com/v1/videos \
-H "Authorization: Bearer $API_NEED_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-h3",
"prompt": "A cinematic product film begins with a quiet tabletop at dawn, then the camera arcs around a glass sculpture while soft wind moves the surrounding fabric. Natural reflections, restrained motion, no text overlay.",
"duration": 5,
"size": "768P",
"metadata": {
"ratio": "16:9"
}
}'Poll the task
Save task_id, then poll GET /v1/videos/{task_id} every few seconds until the task is completed or failed.
curl https://apineed.com/v1/videos/$TASK_ID -H "Authorization: Bearer $API_NEED_API_KEY"MiniMax H3 API Endpoint
Create an asynchronous video task, then use its task ID to retrieve status and the completed video URL.
/v1/videos- Authorization
- Bearer $API_NEED_API_KEY
- Content-Type
- application/json
- HTTP-Referer
- optional - your site URL, for rankings
- Model
- minimax-h3
/v1/videos/{task_id}- Authorization
- Bearer $API_NEED_API_KEY
- Content-Type
- application/json
- HTTP-Referer
- optional - your site URL, for rankings
- Model
- minimax-h3
Parameters
These controls follow APINEED's current DTO and adapter behavior. Unsupported upstream-only fields are intentionally omitted.
| Name | Type | Values | Description |
|---|---|---|---|
model | string | Required | Use minimax-h3. |
prompt | string | Required | Describe the scene, action, camera movement, light, and ending state. |
duration | integer | 4-15 | Whole-second video duration; defaults to 5. |
size | string | 768P, 2K | Selects the validated MiniMax H3 output resolution. |
metadata.ratio | string | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | Sent only for text-to-video requests without reference frames. |
images[0] | string | Optional | Accessible first-frame image URL or data URL. |
images[1] | string | Optional | Accessible final-frame image URL or data URL. |
MiniMax H3 API questions
Practical answers for choosing, integrating, pricing, and operating the MiniMax H3 API on APINEED.
Which MiniMax H3 workflows does APINEED expose?
The current APINEED page exposes text-to-video, one-image first-frame generation, and two-image start-to-end frame generation. It does not expose H3 reference video or reference audio until those payloads are verified end to end. [2][4]
What duration and resolution can I request from MiniMax H3?
Use a whole-number duration from four through fifteen seconds and choose either 768P or 2K. APINEED validates both fields before dispatch so invalid values do not become unbounded billing multipliers. [2][4]
When does APINEED send the MiniMax H3 aspect ratio?
For text-to-video, APINEED sends metadata.ratio from the six visible choices. When one or two frame images are present, it omits ratio and lets the reference frame dimensions define the composition. [2][4]
Can MiniMax H3 generate synchronized audio?
MiniMax describes H3 as a native video-and-audio model. APINEED may return video containing audio, but this page does not expose unsupported voice, reference-audio, or audio-mixing parameters on the current adapter. [1][2][4]
How is MiniMax H3 priced on APINEED?
The TokenDance channel anchor is ¥0.50 per generated second. APINEED applies the requested 0.92 multiplier, producing ¥0.46 per second, shown as about $0.068657 at the configured 6.70 CNY per USD rate. [3][4]
How does an application retrieve a MiniMax H3 result?
Create the task with POST /v1/videos, preserve the returned task_id, and poll GET /v1/videos/{task_id}. A completed task exposes the upstream result URL without requiring the browser to keep the creation request open. [3][4]
Is the video shown before generation a MiniMax H3 quality sample?
No. It is explicitly an interface preview used to verify the player layout and controls. Only the URL returned by a completed H3 task should be treated as output evidence for this model. [4]
How to deploy MiniMax H3
Connect MiniMax H3 to your application through the APINEED media API.
Add credits
Add APINEED credits before submitting a billable generation.
Create an API key
Create and securely store an APINEED API key.
Configure the endpoint
Send minimax-h3 requests to /v1/videos.
Track the task
Poll the returned task ID and store the completed video URL.
