generate_video

Pro required

generate_video requires a valid VIDEO_MCP_LICENSE_KEY and a MINIMAX_API_KEY. See Pro Tier and Environment Variables. Your MiniMax account also needs sufficient balance; top up at MiniMax Balance.

Asynchronous

Video generation is an async operation. This tool returns a task_id immediately - use query_generation_task to poll for completion and download the result.

Generate a video from a text description and/or first/last-frame image using MiniMax. The default model is MiniMax-H3, MiniMax’s next-generation flagship served by the new V2 API. V1 models (T2V-01, T2V-01-Director, MiniMax-Hailuo-2.3, MiniMax-Hailuo-02) remain selectable for cheaper drafts. Video generation typically takes 1–5 minutes.

Parameters

Parameter	Type	Required	Default	Description
prompt	string	✓*	—	Text description of the video (max 2,000 chars). Supports camera commands (see below). Required for MiniMax-H3.
model	string	—	MiniMax-H3	Model: MiniMax-H3 (V2 API), MiniMax-Hailuo-2.3, MiniMax-Hailuo-02, T2V-01-Director, T2V-01 (V1 API)
first_frame_image	string	✓*	—	First frame as a base64 data URI or public URL
last_frame_image	string	—	—	MiniMax-H3 only: last frame image (base64 data URI or public URL). Combines with first_frame_image for first+last frame video. Mutually exclusive with reference inputs.
reference_image_urls	string[]	—	—	MiniMax-H3 only: reference image URLs (max 9) for reference-to-video. Mutually exclusive with first/last frame images.
reference_video_urls	string[]	—	—	MiniMax-H3 only: reference video URLs (max 3) for reference-to-video. Mutually exclusive with first/last frame images.
reference_audio_urls	string[]	—	—	MiniMax-H3 only: reference audio URLs (max 3) for reference-to-video. Mutually exclusive with first/last frame images.
duration	integer	—	5 (H3) / 6 (V1)	Duration in seconds. MiniMax-H3: 4–15 (default 5). V1 models: 6 or 10 (default 6).
resolution	string	—	768P (H3)	Output resolution. V1 models: 720P, 768P, 1080P. MiniMax-H3: 768P (default) or 2K.
ratio	string	—	—	MiniMax-H3 only: adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16. Required (non-adaptive) for text-to-video; always adaptive for image-to-video.
prompt_optimizer	boolean	—	true	Auto-optimize the prompt for better results. V1 models only.

* At least one of prompt or first_frame_image is required. MiniMax-H3 additionally requires a text prompt.

Camera Commands

When using MiniMax-H3 or a Hailuo model (MiniMax-Hailuo-2.3, MiniMax-Hailuo-02), you can embed camera direction commands in the prompt:

Command	Effect
[Pan left]	Camera pans left
[Pan right]	Camera pans right
[Zoom in]	Camera zooms in
[Zoom out]	Camera zooms out
[Tracking shot]	Camera follows the subject
[Aerial shot]	Bird’s-eye overhead view
[Arc shot]	Camera arcs around the subject

Two-Step Workflow

Step 1: generate_video(prompt="...") → returns task_id
Step 2: query_generation_task(task_id="...") → returns status + download URL when complete

Poll every 30–60 seconds until the status is Success. The default api: 'auto' on query_generation_task handles both V1 and V2 (MiniMax-H3) tasks automatically.

Usage Examples

Text-to-video (MiniMax-H3, default):

“A golden retriever playing fetch on a beach at sunset, slow motion.” (add ratio: "16:9" — a non-adaptive ratio is required for text-to-video)

With camera commands and resolution:

“Generate a video of a cat jumping off a diving board [Tracking shot], then splashing into the pool [Zoom in]. Use MiniMax-H3 at 2K, 5 seconds.”

Cheaper V1 draft:

“Generate a 10-second video of a cat jumping off a diving board [Tracking shot], then splashing into the pool [Zoom in]. Use MiniMax-Hailuo-2.3 at 1080P.”

Image-to-video:

“Create a video from this image: [attach image] — the person starts walking forward through a forest.”

Combined pipeline (image → video):

Step 1: generate_image(prompt="A futuristic car on a neon-lit highway, 16:9")
Step 2: generate_video(prompt="The car starts driving forward through the city",
                       first_frame_image="<base64 from step 1>")
Step 3: query_generation_task(task_id="<from step 2>")

Response

Returns a text part with the task_id, model, and API version used, and instructions to poll with query_generation_task.

Pricing

Video generation is billed to your MiniMax account balance (MINIMAX_API_KEY). The default MiniMax-H3 (V2 API) is billed per output second: $0.13/s at 2K and $0.08/s at 768P (e.g. 5 s @ 2K = $0.65, 5 s @ 768P = $0.40). The first 5 reference images are free, then $0.04 each; reference video is billed per input second at the output resolution rate; reference audio is free.

V1 models are billed per video and are cheaper for casual drafts — e.g. MiniMax-Hailuo-2.3 is $0.28 for a 768P, 6 s clip. Select a V1 model explicitly when you want V1 pricing.

See Model Pricing for the full breakdown of video, image, music, and TTS costs.

Notes

• MiniMax-H3 is the current default and highest-quality model (successor to Hailuo 2.3), served by the V2 API at https://api.minimax.io. It is billed per second (Pay-as-You-Go).
• V1 models (T2V-01, T2V-01-Director, MiniMax-Hailuo-2.3, MiniMax-Hailuo-02) are billed per video and are cheaper for drafts — select them explicitly when you want V1 pricing.
• Reference URLs must be publicly accessible URLs or base64 data URIs — the V2 API accepts public URLs and base64 only.
• query_generation_task polls both API generations automatically with api: 'auto' (default) — see query_generation_task.
• Camera commands are supported on MiniMax-H3 and Hailuo models (MiniMax-Hailuo-2.3, MiniMax-Hailuo-02).
• The MINIMAX_REQUEST_TIMEOUT_MS env var controls how long the MCP client waits for the initial API call (not the full generation time).