MiniMax Hailuo 02: Capabilities, Specs, and API Guide

What is MiniMax Hailuo 02?

MiniMax Hailuo 02 is a video-generation model that supports text-to-video and image-to-video. MiniMax’s model guide lists 512p and 768p clips at 6 or 10 seconds, and 1080p clips at 6 seconds. Its API documentation also describes a first-and-last-frame mode. These are specifications for the named Hailuo 02 model—not proof that every third-party listing offers every mode or setting. (MiniMax model guide)

MiniMax groups Hailuo 02 under Legacy Models in its model guide, while its API model list still names MiniMax-Hailuo-02. The docs do not say it has been retired. For a new production integration, confirm that the model remains available to your account and compare it with MiniMax’s currently promoted video models before committing. (MiniMax model guide)

Is Hailuo 02 a fit for your video task?

Use the exact input and output requirements to decide:

• Starting from a prompt: The official API has a text-to-video task route. The model guide lists Hailuo 02 as supporting text-to-video. (MiniMax text-to-video API)
• Animating a supplied image: The image-to-video API requires a first-frame image and allows a text prompt. For this API mode, MiniMax lists JPG, JPEG, PNG, and WebP; an image smaller than 20 MB; a short edge above 300 pixels; and an aspect ratio between 2:5 and 5:2. (MiniMax image-to-video API schema)
• Controlling the ending frame: MiniMax documents first-and-last-frame generation for Hailuo 02. This mode does not support 512p; the API schema lists 6- or 10-second clips at 768p and 6-second clips at 1080p. (MiniMax first-and-last-frame API schema)
• Needing a particular resolution or length: Match your requirement against the model guide’s listed combinations: 512p or 768p at 6 or 10 seconds, or 1080p at 6 seconds. Confirm the mode-specific API schema too, since settings can vary by task type. (MiniMax model guide)

MiniMax describes Hailuo 02 as strong at instruction following and physical realism. WaveSpeedAI also advertises physics-driven motion and prompt responsiveness for its Hailuo 02 I2V Standard listing. These are provider descriptions, not comparative test results; don’t treat them as proof that Hailuo 02 will outperform another model on your footage or prompts. (MiniMax model list; WaveSpeedAI Hailuo 02 I2V Standard listing)

Practical selection rule: choose Hailuo 02 when its documented modality, resolution, and short clip lengths fit the job. If your decision depends on visual quality, character consistency, or reliable camera movement, test representative inputs and judge the generated clips against your own acceptance criteria. Do not rely on an unqualified benchmark-rank claim: a third-party page reports a ranking without enough detail to assess it. (MiniMax image-to-video API)

Hailuo 02 API: a direct MiniMax example

MiniMax’s API documentation uses the MiniMax-Hailuo-02 model name, bearer-token authentication, and POST https://api.minimax.io/v1/video_generation for video-generation tasks. The image-to-video request below shows the documented endpoint and request pattern. Replace the example image URL and <MINIMAX_API_KEY> with an image accessible to the API and your key. (platform.minimax.io)

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-02",
    "first_frame_image": "https://example.com/starting-frame.jpg",
    "prompt": "A red paper kite lifts into the air; slow camera push-in.",
    "duration": 6,
    "resolution": "768P"
  }'

The image-to-video schema requires model and first_frame_image; prompt is optional. MiniMax’s asynchronous task flow uses a returned task_id for status queries. The submission-response excerpt does not specify where task_id is nested, so inspect the response or current reference instead of assuming a JSON path. (MiniMax image-to-video API schema; MiniMax query-task reference)

MiniMax’s query reference documents the GET status endpoint and bearer authentication, but its displayed request example omits the task identifier. The command below is therefore a reference template: confirm the current reference’s parameter placement before running it. The file-retrieval reference likewise shows the GET endpoint and bearer authentication; confirm its current file_id request placement before using that command.

# Reference template: confirm task_id parameter placement in the current API docs.
curl --get 'https://api.minimax.io/v1/query/video_generation' \
  --data-urlencode "task_id=$TASK_ID" \
  --header 'Authorization: Bearer <MINIMAX_API_KEY>'

# Reference template: confirm file_id parameter placement in the current API docs.
curl --get 'https://api.minimax.io/v1/files/retrieve' \
  --data-urlencode "file_id=$FILE_ID" \
  --header 'Authorization: Bearer <MINIMAX_API_KEY>'

The query reference lists Preparing, Queueing, and Processing as in-progress statuses, Success as completion, and Fail as failure. On success, use the returned file_id with the file-retrieval endpoint. The retrieval response example places the download link at file.download_url; download the video from that URL. (MiniMax query-task reference; MiniMax Retrieve File reference)

This is a direct MiniMax API example. It is different from WaveSpeedAI’s API, which uses a WaveSpeedAI key, a WaveSpeedAI endpoint, and a prediction-ID polling route. Don’t substitute one provider’s endpoint, authentication, or response handling for another’s. (WaveSpeedAI Hailuo 02 I2V Standard listing)

Pricing and production checks

The official MiniMax pricing page surfaced here lists per-second rates for H3 models, but not a usable Hailuo 02 rate. A WaveSpeedAI listing gives a different, provider-specific price table: $0.28 per 6-second 768p video and $0.56 per 10-second 768p video. The same listing’s FAQ ambiguously says “$0.28 per run or per second,” and warns that listed prices may be outdated. Treat those figures as WaveSpeedAI listing information—not MiniMax direct API pricing—and check the actual provider’s current estimate and final charge before budgeting. (MiniMax pricing overview; WaveSpeedAI Hailuo 02 I2V Standard listing)

Before integrating, verify:

1. Availability and lifecycle: confirm MiniMax-Hailuo-02 is enabled for your account, especially since MiniMax labels it a legacy model in its model guide. (MiniMax model guide)
2. Exact task schema: validate the mode, image requirements, duration, resolution, and response/query fields in the official API reference for the endpoint you will call. (MiniMax image-to-video API schema)
3. Cost and timing: confirm current provider-specific billing and test end-to-end latency with your own workload. A third-party estimate is not a service-level guarantee. (WaveSpeedAI Hailuo 02 I2V Standard listing)
4. Rights and operational terms: review the applicable MiniMax and platform terms for your intended commercial use. The captured WaveSpeedAI listing does not itself grant commercial rights. (WaveSpeedAI Hailuo 02 I2V Standard listing)
5. Provider route: this article documents MiniMax’s API and separately identifies WaveSpeedAI’s listing. It does not establish Hailuo 02 availability or an endpoint on TokenForge Cloud.

Bottom line: Hailuo 02 is a documented choice for short text- or image-led video generation, with mode-specific resolution and duration limits. Use MiniMax’s own API docs when integrating directly; use a reseller or gateway only after validating that provider’s exact model listing, schema, price, and terms.

Sources

Query Video Generation Task Status - MiniMax API Docs

Retrieve File - MiniMax API Docs

MiniMax Image-to-Video API Schema

MiniMax First-and-Last-Frame API Schema

Models - MiniMax API Docs

Pricing Overview - MiniMax API Docs

Hailuo 02 I2V Standard Image-to-Video API | WaveSpeed