MiniMax (Hailuo) Commands & Prompt Reference
24 commands & parameters · 3 categories · Source: MiniMax Platform API Reference
MiniMax splits its surface in two: video generation (Hailuo) takes a multimodal content array where each item declares a type - text, image_url, video_url, or audio_url - alongside resolution, duration, and ratio; the chat API is OpenAI-compatible and adds its own thinking and reasoning_split controls on top of the familiar sampling fields. This page lists both from the official API reference.
The video endpoint is asynchronous: a successful call returns a task_id and you poll the query endpoint, or set callback_url to be notified. Every video scenario must include at least one text content item - the other content types are references layered on top of it.
All Commands
Video Generation Parameters (6)
| Command / Syntax | What it does | Notes |
|---|---|---|
| model | Model name | Currently available: MiniMax-H3. |
| content | Array of multimodal input describing the information used to generate the video | Each element is distinguished by type (text / image_url / video_url / audio_url) and can be labeled with a role. Every request must include one non-empty text item (the prompt is... |
| resolution | Video resolution | Currently available: 768P, 2K. |
| duration | Duration of the generated video in seconds | Required, integer. Available values: 4-15. |
| ratio | Aspect ratio of the generated video | Defaults to adaptive (the most suitable ratio is chosen automatically based on the input; the actual ratio can be read from the ratio field of the query endpoint). Available values:... |
| callback_url | Callback URL for task status changes | Once configured, the MiniMax server first sends a verification request containing a challenge field (you must return the challenge unchanged within 3 seconds to complete verification);... |
Multimodal Content & Model Tokens (6)
| Command / Syntax | What it does | Notes |
|---|---|---|
| "type": "text" | Text prompt content item; every scenario must include one | Inside the content array, with the prompt in the text field. |
| "type": "image_url" | Image input content item | Supplies a reference or first-frame image to the generation. |
| "type": "video_url" | Video input content item | Supplies a reference video to the generation. |
| "type": "audio_url" | Audio input content item | Supplies reference audio to the generation. |
| role | Position or purpose of a content item | Conditionally required depending on the content type. |
| MiniMax-H3 | Current video generation model ID (Hailuo 03) | Set as the model field value. |
Chat API Parameters (12)
| Command / Syntax | What it does | Notes |
|---|---|---|
| model | Model ID | JSON body field of the OpenAI-compatible chat endpoint. |
| service_tier | Service tier for request admission | Supported values are standard and priority. If omitted, the request uses the standard tier. The priority price is 1.5 times the standard price and ensures priority admission so the... |
| messages | A list of messages containing the conversation history | Supports text, image, video, and tool call content. |
| thinking | Controls MiniMax-M3 thinking | When omitted, adaptive thinking is enabled by default and responses include thinking content. For M2.x models, thinking cannot be disabled. |
| reasoning_split | Output-format switch | When enabled, separates thinking content into the reasoning_content and reasoning_details fields. This does not enable or disable thinking. |
| stream | Whether to use streaming output, defaults to false | When set to true, the response will be returned in chunks. |
| stream_options | Streaming response options | JSON body field of the OpenAI-compatible chat endpoint. |
| max_completion_tokens | Specifies the upper limit for generated content length, in tokens | For MiniMax-M3 the recommended value is 131072 (128K) and the maximum is 524288 (512K); for other models the recommended value is 65536 (64K) and the maximum is 204800 (200K). If... |
| temperature | Temperature coefficient, affects output randomness | Range [0, 2], default 1. Higher values produce more random output; lower values produce more deterministic output. |
| top_p | Nucleus sampling parameter | Range [0, 1]. Default is 0.95 for MiniMax-M3 and 0.9 for M2.x models. |
| tools | Tool definition list | Function tools are supported. |
| max_tokens | Legacy generation length limit parameter | Deprecated; use max_completion_tokens instead. |
Recipes
Our own notes on combining the parameters and commands above. Every token used here appears in the reference tables on this page.
Shape a video request
MiniMax-H3 + content + resolution / duration / ratio
The content array mixes items by type - every scenario must include one "type": "text" item, with "type": "image_url" or "type": "video_url" supplying the visual reference - while resolution, duration, and ratio pin the output format.
Get told when the render finishes
callback_url
Configure a callback and MiniMax first sends a verification request with a challenge field you must echo back unchanged; after that, task status changes arrive as callbacks instead of polling.
Separate the thinking from the answer
thinking + reasoning_split
Adaptive thinking is on by default and thinking content arrives inline. reasoning_split moves it into reasoning_content and reasoning_details fields without turning thinking on or off - on M2.x models it cannot be disabled at all.
Pay for priority only when it matters
service_tier
standard is the default; priority costs 1.5 times the standard price. Set it per request, so a latency-sensitive call can jump the queue while batch work stays on standard.
Frequently Asked Questions
How do I pass a prompt to MiniMax video generation?
As a content item with type text. The reference is explicit that every scenario must include one text item; image, video, and audio items are references added alongside it.
Which model IDs are current?
MiniMax-H3 is the listed video generation model (Hailuo 03). Chat model IDs are set through the model field on the OpenAI-compatible endpoint.
Is video generation synchronous?
No. The endpoint returns a task_id; poll the query task endpoint or configure callback_url for status notifications.
Every entry on this page is a literal, typeable token taken from the official documentation linked above. Command sets in AI tools change quickly - the in-product command menu is always authoritative for your installed version.