MiniMax Video API Overview
Call H3, Context-IR, Max, Max Turbo, and video regeneration through aiiqq.com with the correct creation and query formats.
MiniMax tasks run asynchronously: submit a request, save its public task ID, poll for a terminal state, then read the video URL or enhanced prompt. This series describes the interfaces integrated into aiiqq.com and references the official MiniMax Video V2 API.
Supported models and operations
| aiiqq.com model name | Purpose | Creation path | Request-body model |
|---|---|---|---|
minimax-h3 | Generate 768P / 2K video | /v2/video_generation | MiniMax-H3 |
minimax-h3-context-ir | Understand media and enhance a prompt; returns text | /v2/h3_context_ir | MiniMax-H3 |
minimax-h3-regeneration | Regenerate an eligible H3 768P video at 2K | /v2/video_regeneration | MiniMax-H3 |
minimax-h3-max | Generate 480P / 768P video | /v1/video/generate | minimax-h3-max |
minimax-h3-max-turbo | Platform-format video generation | /v1/video/generate | minimax-h3-max-turbo |
The first three names identify permissions and billing. All three native request bodies use MiniMax-H3; the creation path selects the operation. Max and Max Turbo use lowercase names and the separate platform format. Turbo-specific media and extended options depend on current model support; do not assume every Max option is available.
Base URL and authentication
export MODELSELL_BASE_URL="https://api.modelsell.com"
export MODELSELL_API_KEY="<your-modelsell-api-key>"Send requests to aiiqq.com with Authorization: Bearer $MODELSELL_API_KEY; creation requests use Content-Type: application/json. Allow at least 300 seconds for a request. For your own gateway domain, replace only MODELSELL_BASE_URL.
Choose an interface
| Format | Create | Query | Successful result |
|---|---|---|---|
| H3 native | The three /v2 creation paths above | GET /v2/query/video_generation/{task_id} | task.content.url or task.content.prompt |
| Max / Turbo platform | POST /v1/video/generate | GET /v1/video/tasks/{task_id} | task.outputs |
| Generic video | POST /v1/videos | GET /v1/videos/{task_id} | video_url / metadata.url |
The generic format supports H3, Max, and Max Turbo generation; see Generic Video Format. Use the dedicated native creation paths for Context-IR and regeneration.
The two native query paths are not interchangeable. A returned task_… is a ModelSell public ID; do not replace it with a numeric official example ID, an upstream ID, or an ID from a callback. Native queries, deletion, and source references check the owning user and creation token, and reuse the original channel and supplier key. The examples retain the same aiiqq.com API Key used to create the task.
Native H3 generation
curl --max-time 300 "$MODELSELL_BASE_URL/v2/video_generation" \
-H "Authorization: Bearer $MODELSELL_API_KEY" \
-H 'Content-Type: application/json' \
-d '{
"model": "MiniMax-H3",
"content": [{"type": "text", "text": "A red ceramic teapot rests on a wooden table in warm morning sunlight. The camera moves slowly closer."}],
"resolution": "768P",
"duration": 4,
"ratio": "16:9",
"aigc_watermark": false
}'Example creation response:
{"task_id": "task_example_h3_generation"}Poll every 10–15 seconds with the actual public ID returned by your request:
export TASK_ID="<returned-task-id>"
curl --max-time 300 "$MODELSELL_BASE_URL/v2/query/video_generation/$TASK_ID" \
-H "Authorization: Bearer $MODELSELL_API_KEY"Prompt enhancement and regeneration
Send model: "MiniMax-H3", content, duration, and the applicable ratio to POST /v2/h3_context_ir. On success, read task.content.prompt. This is enhanced text rather than a video file; use it as the text input of a subsequent H3 generation request.
Regeneration can reference a successful H3 generation task created by the same token within the last seven days:
{
"model": "MiniMax-H3",
"source_task_id": "task_example_successful_h3",
"resolution": "2K"
}Send this JSON to POST /v2/video_regeneration. Use the aiiqq.com public ID for source_task_id. The source must use the same native protocol and original channel key; Max / Turbo platform tasks cannot be referenced. A successful H3 task created through the generic format and served by the H3 native protocol is also eligible. Alternatively, provide an eligible base_video; see Request Parameters and Media.
States and results
| Format | Waiting / processing | Success | Failure terminal states |
|---|---|---|---|
H3 native task.status | queued / running | succeeded | failed / cancelled |
Platform task.status | pending / processing | completed | failed |
Generic video status | queued / in_progress | completed | failed |
Download native videos from task.content.url; read Context-IR text from task.content.prompt. Platform video URLs are in task.outputs, and metering is in task.metadata.usage; task.usage does not necessarily contain seconds. Result URLs may expire, so download or store outputs promptly. The Python example provides bounded polling, failure handling, and downloading.
Cancellation, deletion, and errors
Use DELETE /v2/video_generation/{task_id} for H3 native tasks. Queued tasks can be cancelled; running tasks are rejected; successful or failed tasks can be deleted. Cancellation leaves a queryable cancelled state, while deletion invalidates the public ID. Deleting a completed result does not refund its generation fee.
Completed tasks are queried and settled before deletion. If a 409 indicates settlement is still pending, query and retry later. If final usage became unavailable after deletion, the task enters RECONCILING and retains the reservation for reconciliation; do not repeatedly delete it or treat it as refunded. This series provides no platform Max / Turbo cancel, delete, or task-list endpoints.
Registered H3 native endpoints use MiniMax-style errors:
{
"type": "error",
"error": {"type": "bad_request_error", "message": "invalid params", "http_code": "400"},
"request_id": "request_example"
}Inspect the HTTP status and error.message, and retain request_id. Unknown paths may use the gateway's generic error shape; platform requests retain the platform error format. If creation times out, retain diagnostic information and establish whether a task was accepted before submitting again.
Billing and availability
Creation reserves quota. Successful tasks settle against actual usage; failed or cancelled tasks follow the task refund flow. Queries do not incur a second charge.
| Operation | Final metering |
|---|---|
| H3 / regeneration | task.usage.total_seconds, including applicable input-video seconds; the first five reference images are free |
| Context-IR | task.usage.prompt_tokens and completion_tokens, priced separately |
| Max | Separate task.metadata.usage.output_seconds and input_seconds; the first two reference images are free |
| Max Turbo | The video-second rule published for the current model |
See ModelSell MiniMax H3 pricing for current unit and group prices.
Official documentation can describe additional products. This series currently excludes Highspeed, Lite, Max on /v2, native task lists, and MiniMax file APIs. Use public HTTP(S) media URLs rather than mm_file://. The upstream provider delivers callback_url callbacks and may include its own task ID; clients must retain the aiiqq.com public ID from creation.