Text to video
Send a prompt with MiniMax-Hailuo-2.3 to create a video directly from written scene and motion direction.
MiniMax H3 is the name of this independent product experience, not an official API model. The integration details on this page map transparently to MiniMax-Hailuo-2.3 and the other public video models documented by MiniMax.
curl --request POST \
--url https://api.minimax.io/v1/video_generation \
--header "Authorization: Bearer $MINIMAX_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"model": "MiniMax-Hailuo-2.3",
"prompt": "A cyclist crosses a rain-lit city street [Tracking shot], then looks toward camera [Push in].",
"duration": 6,
"resolution": "1080P",
"prompt_optimizer": true
}'The public API uses a shared generation endpoint. The selected model and supplied media determine which workflow is being requested.
Send a prompt with MiniMax-Hailuo-2.3 to create a video directly from written scene and motion direction.
Add first_frame_image to animate a public URL or supported Base64 Data URL from a controlled opening frame.
Supply first_frame_image and last_frame_image to describe the transition between two controlled images.
Use the separate S2V-01 workflow when facial identity consistency from a character reference is the primary goal.
Generation does not keep one HTTP request open. Store the returned identifiers and move through the three documented API steps.
Send the model, prompt, duration, resolution, and any supported media inputs. A successful response returns task_id.
POST /v1/video_generationQuery the task until it reaches Success or Fail. Production backends can also provide callback_url and complete MiniMax’s validation handshake.
GET /v1/query/video_generation?task_id=...When the task succeeds, exchange file_id for file metadata and a download_url, then copy the result into storage you control if it must remain durable.
GET /v1/files/retrieve?file_id=...The MiniMax video API family contains multiple public models. Capabilities should be attributed to the specific model rather than to the independent H3 product name.
| Model ID | Primary workflow | 6-second output | 10-second output |
|---|---|---|---|
| MiniMax-Hailuo-2.3 | Text to video · Image to video | 768P or 1080P | 768P |
| MiniMax-Hailuo-2.3-Fast | Image to video | 768P or 1080P | 768P |
| MiniMax-Hailuo-02 | Text · Image · First/last frame | 768P or 1080P | 768P |
| S2V-01 | Character subject reference | Model-specific output | Not documented |
Model availability, rate limits, pricing, and supported combinations can change. Validate against the current official API reference before shipping.
Keep the public request explicit and validate model-specific combinations before a task is submitted.
Use the exact documented identifier. MiniMax-Hailuo-2.3 supports text and image generation; the Fast variant is image-to-video.
Describe subject, scene, motion, camera behavior, and atmosphere. Supported models also accept bracketed camera commands.
Hailuo 2.3 supports 6 or 10 seconds at 768P and 6 seconds at 1080P.
Use a combination supported by the selected model and duration. 1080P is a 6-second option for Hailuo 2.3.
Leave optimization enabled for broad creative interpretation, or disable it when exact prompt and camera wording must be preserved.
Provide a public callback endpoint, echo the validation challenge within three seconds, and verify updates before trusting them.
Polling clients should tolerate every documented intermediate state and stop only on the two terminal outcomes.
| Status | Meaning | Recommended application behavior |
|---|---|---|
| Preparing | The task is being prepared | Persist task_id and continue waiting |
| Queueing | The task is waiting for capacity | Back off between status checks |
| Processing | Video generation is in progress | Keep the job visible and avoid duplicate submission |
| Success | Generation completed | Read file_id and retrieve the output |
| Fail | Generation did not complete | Show the returned error context and allow a deliberate retry |
The documented image rules are simple enough to enforce at upload time, preventing avoidable task failures.
Use a public URL or a correctly formed Base64 Data URL for first-frame and last-frame image inputs.
Reject oversized uploads before encoding or forwarding them to the generation service.
The documented aspect-ratio range is between 2:5 and 5:2. Avoid extremely narrow source frames.
The first frame determines output geometry. A differently sized last frame may be cropped to match.
Official prices and packages vary by region and access channel. Link to MiniMax’s live pricing documentation instead of hard-coding a number into your integration.
The most important distinction is between this site’s product name and MiniMax’s public API model names.
No public MiniMax video API model named MiniMax-H3 is listed in the official documentation. This independent site uses MiniMax H3 as a product label; API requests should use a documented ID such as MiniMax-Hailuo-2.3.
No. The current generator is a browser-only simulation. It does not create MiniMax tasks, upload reference files, charge an account, or return generated videos.
The current official text-to-video reference documents callback_url. Your endpoint must complete the validation challenge and should authenticate and de-duplicate subsequent status updates.
A successful status query returns file_id. Pass that identifier to GET /v1/files/retrieve to receive file metadata including download_url.
Use the official MiniMax pricing and rate-limit pages for the account region you will deploy against. Those values can change independently of this guide.
Use the MiniMax H3 interface to shape a creative direction, then implement production requests with the current MiniMax API documentation.