How to Use MiniMax H3 Max Turbo for Text and Image to Video in Cursor?

To use MiniMax H3 Max Turbo, add the Makernice MCP server to Claude Code or Cursor and describe the clip, or hand over a still. The agent prices each request with estimate_cost, you say yes, and generate creates the job under the max_cost_usd you approved. At the default 768p, a 5-second clip costs $0.24 on price list v15 (November 12, 2026).

Is there anything to install besides the Makernice server?

No. The MCP server is the only piece. In Claude Code, run claude mcp add --transport http --scope user makernice using the URL from the Claude Code connection steps, then /mcp and a browser sign-in. In Cursor, follow the Cursor connection steps: one-click install or an entry in ~/.cursor/mcp.json, then sign in next to makernice in the MCP settings. No API key is copied anywhere; the editor holds a signed-in connection with its own spending limits.

Because Max Turbo supports fewer modes than its siblings, it is worth having the agent read the model's details first:

{
  "name": "list_models",
  "arguments": {
    "model": "minimax-h3-max-turbo"
  }
}

The answer lists each mode with its resolutions and per-second prices, the accepted aspect ratios and the duration range, so the agent writes a valid request on the first try.

Can I make MiniMax H3 Max Turbo clips in Claude or ChatGPT instead?

Yes, and only the connection differs. Once Makernice is added, the chat app calls estimate_cost, waits for your go-ahead, sends generate with that figure as max_cost_usd and reads the result from check_job, the same three calls Claude Code and Cursor make for MiniMax H3 Max Turbo. The estimate is not a reserved price: generate quotes again and stops if the new quote is over your limit.

• Claude: Customize → Connectors → + Add → Add custom connector, paste the MCP URL, then Connect and sign in. On Team and Enterprise plans an owner adds it in Organization settings first. Claude steps.
• ChatGPT on the web: Plugins → + → Add custom MCP server, Authentication set to OAuth and the ChatGPT-only URL, then type @ in a new chat and pick Makernice. Your plan or workspace must allow custom MCP servers, and that version has no get_topup_link tool, so top up on the site. ChatGPT steps.

How do I write a text-to-video request for MiniMax H3 Max Turbo, and which settings matter?

A MiniMax H3 Max Turbo text-to-video request is mode t2v plus a prompt of up to 4,000 characters, and three settings shape it: aspect_ratio (21:9, 16:9, 4:3, 1:1, 3:4, 9:16), resolution (480p, 768p, 1080p, 768p when left out) and duration (5 to 15 whole seconds, 5 when left out). Only resolution and duration change the price. A vertical drone shot, from request to file:

You: Make a 9:16 drone shot rising over rice terraces at sunrise with MiniMax H3 Max Turbo. How much?

Cursor sends the request to estimate_cost at the default resolution:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "minimax-h3-max-turbo",
      "mode": "t2v",
      "prompt": "Drone rises over green rice terraces at sunrise, mist sits in the valleys, a farmer's hut with smoke from the chimney, slow upward reveal",
      "duration": 5,
      "resolution": "768p",
      "aspect_ratio": "9:16"
    }
  }
}

Cursor: The estimate is $0.24 for 5 seconds at 768p (price list v15). Nothing charged. Start it?

You: Start it.

Cursor reuses the input in generate, with the approved amount and a request_id:

{
  "name": "generate",
  "arguments": {
    "input": {
      "model": "minimax-h3-max-turbo",
      "mode": "t2v",
      "prompt": "Drone rises over green rice terraces at sunrise, mist sits in the valleys, a farmer's hut with smoke from the chimney, slow upward reveal",
      "duration": 5,
      "resolution": "768p",
      "aspect_ratio": "9:16"
    },
    "max_cost_usd": "0.24",
    "request_id": "terraces-sunrise-vertical-01"
  }
}

Makernice quotes again when generate is called and creates the job only if that quote is no more than max_cost_usd and within the connection's per-job limit. You pay the quote, nothing more. If Cursor has to resend because the reply never arrived, the same request_id with the same input brings back the job already created.

Next comes check_job, with wait_s set as high as 55 seconds if the agent should wait. The succeeded job's outputs[0].url is a signed MP4 link valid for 24 hours, renewable with another check_job, and console_url leads to the console record. An account webhook also reports MCP jobs, if you'd rather be notified than wait.

Can MiniMax H3 Max Turbo turn an image into a video?

Yes, MiniMax H3 Max Turbo turns an image into a video in mode i2v: put the still in image_url as a public https link and describe the motion in prompt. Image-to-video on MiniMax H3 Max Turbo has no aspect_ratio field, so the clip keeps the shape of the still; a request that sends one is refused. Duration and resolution follow the same ranges as text-to-video, and the per-second price is the same in every mode:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "minimax-h3-max-turbo",
      "mode": "i2v",
      "prompt": "Mist drifts slowly down the terraces while the camera tilts up toward the rising sun",
      "image_url": "https://example.com/stills/terraces-dawn.jpg",
      "duration": 5,
      "resolution": "768p"
    }
  }
}

Makernice doesn't measure the still for MiniMax H3 Max Turbo before the job starts, so an image the model can't use shows up as a failed job with its hold released. Crop extreme shapes such as tall phone screenshots before sending them.

Modes, resolutions and limits for MiniMax H3 Max Turbo

MiniMax H3 Max Turbo covers text-to-video, image-to-video and first and last frame. It has no reference-to-video mode; for consistent characters from up to 4 reference images, use MiniMax H3 Max. Resolutions are 480p, 768p, 1080p, 768p if you don't choose; text-to-video takes 21:9, 16:9, 4:3, 1:1, 3:4, 9:16. Durations run from 5 to 15 seconds.

Per-second prices on price list v15 (November 12, 2026) are $0.03 at 480p, $0.048 at 768p and $0.096 at 1080p. A 10-second clip costs ten times the per-second figure for its tier; the agent can confirm the total with estimate_cost, as in this first and last frame request:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "minimax-h3-max-turbo",
      "mode": "flf",
      "prompt": "Time passes over the terraces, the mist burns off and the light turns harsh and bright",
      "image_url": "https://example.com/frames/terraces-dawn.jpg",
      "last_image_url": "https://example.com/frames/terraces-noon.jpg",
      "duration": 10,
      "resolution": "480p"
    }
  }
}

Images are public https links to JPEG, PNG or WebP files of at most 10 MB; local paths are not readable over MCP. The same applies to both frames in first and last frame mode. The MiniMax H3 Max Turbo model page has the full specification.

Reading a refused MiniMax H3 Max Turbo request

Makernice refuses over MCP with a readable message, and a refused call has neither created a job nor charged anything. Column one is the refusal's internal server name; column two is how the returned text starts.