MiniMax-H3 Video Generation (V2) - Overview

Version: v1.0.0 | Last Updated: 2026-08-13
This platform fully supports the MiniMax Video Generation V2 (MiniMax-H3) official API. Requests and responses are transparently proxied, and all parameter semantics remain consistent with the official API.

Capabilities

MiniMax-H3 is MiniMax's next-generation video model. Tasks are created with a multimodal content array (text / image / video / audio), support 768P / 2K native output and 4-15 second durations, and come with two companion capabilities: 768P-to-2K regeneration and H3-Context-IR prompt enhancement.

Capability	Documentation	Description
Video Generation	01-Create-Video-Generation	Text-to-video, image-to-video (first / last / first+last frame), multimodal-reference-to-video
Video Regeneration	02-Create-Video-Regeneration	Regenerate an H3 768P source video into 2K
H3-Context-IR	03-Create-H3-Context-IR	Understand multimodal context and return a structured, enhanced prompt (no video is generated)

Relationship to Hailuo V1: MiniMax-H3 uses a standalone V2 API (route prefix /hailuo/v2/*). Its request / response structures are completely different from the Hailuo V1 API (/hailuo/v1/*); do not mix them.

Endpoints

Endpoint	Method	URL
Create Video Generation Task	`POST`	`https://platform.dataeyes.ai/hailuo/v2/video_generation`
Create Video Regeneration Task	`POST`	`https://platform.dataeyes.ai/hailuo/v2/video_regeneration`
Create H3-Context-IR Task	`POST`	`https://platform.dataeyes.ai/hailuo/v2/h3_context_ir`
Query Task	`GET`	`https://platform.dataeyes.ai/hailuo/v2/query/video_generation/{task_id}`
List Tasks	`GET`	`https://platform.dataeyes.ai/hailuo/v2/query/video_generation`
Cancel or Delete Task	`DELETE`	`https://platform.dataeyes.ai/hailuo/v2/video_generation/{task_id}`

URL Pattern: Append the channel prefix /hailuo after the platform domain, followed by the official V2 API path.
For example, the official https://api.minimaxi.com/v2/video_generation maps to https://platform.dataeyes.ai/hailuo/v2/video_generation.

Standard Workflow

Create Task (POST /hailuo/v2/video_generation | video_regeneration | h3_context_ir)
        │  Returns task_id
        ▼
Query Task (GET /hailuo/v2/query/video_generation/{task_id})   ← Poll until succeeded
        │  Returns content.url (video tasks) or content.prompt (Context-IR tasks)
        ▼
Download / persist the output (download URLs are time-limited — save promptly)

Authentication

All requests require a Bearer token in the HTTP Authorization header:

Authorization: Bearer {API_KEY}

{API_KEY} is the API key created in the DataEyes console.

Request Headers

Header	Required	Description
`Authorization`	Yes	`Bearer {API_KEY}`
`Content-Type`	Yes	Must be `application/json` (for task creation endpoints)

Task Status

Status	Description
`queued`	Waiting in queue; can be cancelled
`running`	Processing
`succeeded`	Completed; output available in `content`
`failed`	Failed; the `error` field contains the code and reason
`cancelled`	Cancelled

Billing & Settlement

Billing dimensions:

Item	How it is billed
Video generation (768P / 2K, unit price by resolution)	Per second, based on `usage.total_seconds` (input reference video seconds + output seconds, same rate)
Video regeneration (always 2K)	Per second, based on the source video's actual duration (`usage.total_seconds`)
Extra input images	First 5 images per task are free; additional images are billed per image
H3-Context-IR	Per token: input billed on `usage.prompt_tokens`, output on `usage.completion_tokens`

Settlement:

Video generation / regeneration: an estimated output fee is pre-charged at submission (generation uses the requested duration; regeneration uses a 5-second estimate). After the task succeeds, the fee is settled against the actual usage returned by the query endpoint — the difference is charged or refunded accordingly.

H3-Context-IR: nothing is charged at submission; token usage is settled after the task succeeds.

Failed tasks: not billed; any pre-charged amount is fully refunded.

Cancelling a queued task: the pre-charged amount is refunded. Deleting a completed task does not affect settled charges.

For unit prices, please refer to the platform pricing page and your commercial agreement.

Error Responses

On error, the HTTP status code is the real error code, and the response body follows an OpenAI-style structure:

{
  "type": "error",
  "error": {
    "type": "bad_request_error",
    "message": "invalid params, content must include a non-empty text item (prompt is required) (2013)",
    "http_code": "400"
  },
  "request_id": "021785229015510a2c883cf675b9804d"
}

The number in parentheses at the end of error.message is the internal error code. Common errors:

HTTP	error.type	Internal Code	Description
400	`bad_request_error`	2013	Invalid parameters (e.g. content missing a non-empty text item)
401	`authorized_error`	1004 / 2049	Authentication failed / invalid API key
402	`insufficient_balance_error`	1008	Insufficient balance or quota
422	`unprocessable_entity_error`	1026	Input contains sensitive content
429	`rate_limit_error`	1002	Rate limited; retry later
500	`server_error`	1000	Server error

04-Query-Task · 05-List-Tasks · 06-Cancel-or-Delete-Task

00-Overview

MiniMax-H3 Video Generation (V2) - Overview#

Capabilities#

Endpoints#

Standard Workflow#

Authentication#

Request Headers#

Task Status#

Billing & Settlement#

Error Responses#

Related Documentation#