minimax-h3

MiniMax Hailuo omni-modal video model with native stereo audio, producing 5-15 second clips at up to 2K. Supports text-to-video, first/last frame transitions and multi-asset references for character and scene consistency.

Overview

Field	Value
Model ID	`minimax-h3`
CLI	`dlazy minimax-h3`
MCP	`minimax-h3` (Claude Code surfaces this as `mcp__dlazy__minimax-h3`)
Type	`video`
Execution	Async task; the CLI polls until completion (`--no-wait` returns `generateId` immediately)
Batch	Supports `--batch <n>` parallel fan-out

Parameters

Arg	Type	Required	Notes
`prompt`	string	Yes	Prompt
`generation_mode`	"text" \| "frames" \| "components"	No	Generation Mode（text=Text to Video; frames=Frames; components=Components）; default `"text"`
`firstFrame`	url	No	First Frame; image (accepts URL, local path, or data: URL); only when generation_mode="frames"
`lastFrame`	url	No	Last Frame; image (accepts URL, local path, or data: URL); only when generation_mode="frames"
`images`	array<url>	No	Images; image (accepts URL, local path, or data: URL); max 12 items; only when generation_mode="components"
`videos`	array<url>	No	Videos; video (accepts URL, local path, or data: URL); max 6 items; only when generation_mode="components"
`audios`	array<url>	No	Audios; audio (accepts URL, local path, or data: URL); max 3 items; only when generation_mode="components"
`resolution`	"480P" \| "768P" \| "2K" \| "4K"	No	Resolution; default `"2K"`
`aspect_ratio`	"21:9" \| "16:9" \| "4:3" \| "1:1" \| "3:4" \| "9:16"	No	Aspect Ratio; default `"16:9"`; only when !(generation_mode="frames")
`duration`	"5" \| … \| "15"	No	Duration (s); default `"5"`

--input @file.json or --input '{...}' can supply all args at once; flags take precedence over --input keys.

CLI Examples

dlazy minimax-h3 --help
dlazy minimax-h3 --prompt "Write your prompt here" --resolution "2K" --duration "5"
dlazy minimax-h3 --prompt "Write your prompt here" --resolution "2K" --duration "5" --dry-run
dlazy minimax-h3 --prompt "Write your prompt here" --resolution "2K" --duration "5" --no-wait
dlazy minimax-h3 --prompt "Write your prompt here" --resolution "2K" --duration "5" --batch 4

Compose with a pipeline

dlazy gpt-image-2 --prompt "reference visual" \
  | dlazy minimax-h3 --generation_mode "components" --images - --prompt "Write your prompt here"

MCP

MCP is consumed by AI clients, not handwritten. Once dLazy is registered as an MCP server the tool appears in the client's tool list — Claude Code surfaces it as mcp__dlazy__minimax-h3; generic clients (e.g. OpenClaw) call it as minimax-h3.

See MCP setup for how to add the server.

Output

{
  "outputs": [
    { "type": "video", "id": "o_...", "url": "https://files.dlazy.com/result.mp4", "mimeType": "video/mp4" }
  ]
}

Media tools emit image / video / audio / file outputs. Use --output url to print only the URLs on stdout.