MiniMax H3 AI Video Generator API documentation
Generate videos with MiniMax H3 AI Video Generator (MiniMax). Supports both text and image input modes. The reference mode accepts image, video, audio inputs.
Endpoints
| Contract item | Value |
|---|---|
| Capability | minimax-h3-ai-video-generator |
| Submit | POST https://vidmage.ai/api/v1/minimax-h3-ai-video-generator/submit |
| Query | POST https://vidmage.ai/api/v1/minimax-h3-ai-video-generator/query |
| Task identifier | taskId |
| Result field | videoUrl |
Authenticate with Authorization: Bearer YOUR_API_KEY. Save the task identifier and query the same capability. Use a stable Idempotency-Key for submission retries.
Product guide: MiniMax H3 API
Parameters and input rules
| Field | Type | Required | Default | Meaning and limits |
|---|---|---|---|---|
prompt | string | Yes | Not specified | Text prompt describing the video to generate (max 20480 chars). Minimum characters: 1Maximum characters: 20480 |
imageUrl | string | No | Not specified | Optional input image URL. When provided, the task runs in image-to-video mode. In that mode, output ratio is inferred from the input image. |
imageUrls | string[] | No | Not specified | Optional multimodal reference images. Public URLs only; 30 MB per file. Maximum items: 9 |
videoUrls | string[] | No | Not specified | Optional multimodal reference videos. Each video is 50 MB max. Maximum items: 3 |
audioUrls | string[] | No | Not specified | Optional multimodal reference audio. Each audio file is 15 MB max. Maximum items: 3 |
lastFrameUrl | string | No | Not specified | Optional public URL for the final frame in image-to-video mode. |
duration | string | No | "5" | Video duration in seconds. Values: "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15" |
aspectRatio | string | No | "16:9" | Output aspect ratio for modes that expose a ratio parameter. Ignored when imageUrl selects a mode whose ratio is inferred from the input image. Values: "16:9", "9:16", "1:1", "4:3", "3:4", "21:9", "adaptive" |
resolution | string | No | "2k" | Output resolution tier. Values: "768p", "2k" |
aigcWatermark | boolean | No | false | Add an AIGC watermark to the generated video. |
text to video
These are public VidMage field names. The server maps them to provider fields. Input media selects the mode; use its limits together with the parameter table.
| Control | Mode-specific rule |
|---|---|
prompt | Required; 1–20480 characters. |
duration | JSON string. Values: "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15". Default: 5. |
aspectRatio | Values: "16:9", "9:16", "1:1", "4:3", "3:4", "21:9". Default: 16:9. |
resolution | Values: "768p", "2k". Default: 2k. |
aigcWatermark | Add an AIGC watermark to the generated video. Default: false |
image to video
These are public VidMage field names. The server maps them to provider fields. Input media selects the mode; use its limits together with the parameter table.
| Control | Mode-specific rule |
|---|---|
prompt | Required; 1–20480 characters. |
duration | JSON string. Values: "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15". Default: 5. |
aspectRatio | Not used in this mode. |
resolution | Values: "768p", "2k". Default: 2k. |
imageUrl | Formats: jpg, jpeg, png, webp. |
lastFrameUrl | Optional final-frame image. |
aigcWatermark | Add an AIGC watermark to the generated video. Default: false |
multimodal video
These are public VidMage field names. The server maps them to provider fields. Input media selects the mode; use its limits together with the parameter table.
| Control | Mode-specific rule |
|---|---|
prompt | Required; 1–20480 characters. |
duration | JSON string. Values: "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15". Default: 5. |
aspectRatio | Values: "adaptive", "16:9", "9:16", "1:1", "4:3", "3:4", "21:9". Default: adaptive. |
resolution | Values: "768p", "2k". Default: 2k. |
imageUrls | Up to 9 files. Maximum 30 MB per file. Formats: jpg, jpeg, png, webp. |
videoUrls | Up to 3 files. Maximum 50 MB per file. Formats: mp4, mov. |
audioUrls | Up to 3 files. Maximum 15 MB per file. Formats: mp3, wav. |
aigcWatermark | Add an AIGC watermark to the generated video. Default: false |
Example request
{
"prompt": "A native 2K product film where the reference image controls the object, the reference clip controls the camera move, and the final frame lands on a clean hero composition",
"duration": "5",
"aspectRatio": "16:9",
"resolution": "2k",
"aigcWatermark": false
}# Set VIDMAGE_IDEMPOTENCY_KEY to a unique value for this task; preserve it for transport retries.
curl -X POST "https://vidmage.ai/api/v1/minimax-h3-ai-video-generator/submit" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Idempotency-Key: ${VIDMAGE_IDEMPOTENCY_KEY}" \
-H "Content-Type: application/json" \
-d '{"prompt":"A native 2K product film where the reference image controls the object, the reference clip controls the camera move, and the final frame lands on a clean hero composition","duration":"5","aspectRatio":"16:9","resolution":"2k","aigcWatermark":false}'
# -> { "success": true, "taskId": "...", "creditsConsumed": ... }Task results and recovery
URL of the generated video. Read videoUrl from the completed query response. Preserve taskId while the task is running.
curl -X POST "https://vidmage.ai/api/v1/minimax-h3-ai-video-generator/query" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "taskId": "TASK_ID_FROM_SUBMIT" }'
# -> { "success": true, "data": { "status": "...", "videoUrl": "https://..." } }Submission and task queries report the available task and usage information. A timeout is not a confirmed failure. Recover the existing task instead of resubmitting.
For face selection and error recovery, follow Task lifecycle. MCP uses the normalized taskId argument in get_task_result, including capabilities whose REST identifier is requestId.
Credits
| Component / option | Rate | Minimum |
|---|---|---|
| 768p | 15 credits / second | — |
| 2k | 25 credits / second | — |
Base charge = duration × the selected resolution rate.
Each reference image beyond 5 adds 10 credits.
Use the Playground or the MCP credit estimator for your exact inputs. Credits and billing.
