How to Use MiniMax H3-Image for Pictures and Edits in Claude Code?

To use MiniMax H3-Image in Claude Code or Cursor, connect the Makernice MCP server and ask for a picture from text or an edit built on up to 16 reference photos. The agent requests a quote with estimate_cost, gets your approval and then calls generate. A 1k image from text is $0.036 at price list v15 (November 12, 2026).

Connection steps for Claude Code and Cursor

Claude Code: go to the Claude Code setup, copy the claude mcp add --transport http --scope user makernice line including the server address, and run it. Start a session, enter /mcp, select makernice and log in to Makernice when the browser opens.

Cursor: the Cursor setup installs the server with one click or gives you the ~/.cursor/mcp.json entry. Then log in next to makernice in Cursor's MCP settings. Once connected, get_balance reports your balance, the connection's per-job limit and any daily cap.

MiniMax H3-Image without an editor: Claude and ChatGPT connectors

If you would rather ask for MiniMax H3-Image in a chat window than in an IDE, Claude and ChatGPT can reach the same Makernice tools. Claude takes the server as a custom connector (how to add it). ChatGPT on the web takes it as a custom MCP server with OAuth sign-in, on plans or workspaces that permit those, and points at its own server address (how to add it); that version leaves out get_topup_link. Nothing else changes for MiniMax H3-Image: the assistant gets a quote from estimate_cost, which does not reserve the amount, passes the figure you approve to generate as max_cost_usd, and fetches the finished image with check_job.

Can MiniMax H3-Image turn an image into a video?

No, image-to-video is not supported on MiniMax H3-Image: the model makes still images only. To animate a MiniMax H3-Image picture, pass its link as image_url to MiniMax H3 or MiniMax H3 Max in mode i2v; the MiniMax H3 guide walks through that request.

The two names are close, so watch the model id: MiniMax H3 is a video model whose id is minimax-h3. Tell the agent you want a still image, or name the id minimax-h3-image outright. The video id won't accept an image mode, and an image request needs mode t2i or i2i. If in doubt, the agent can call list_models without arguments and read the type field, which says video or image for every model.

A 21:9 banner from MiniMax H3-Image, estimate first

You: We need an ultra-wide homepage banner of an alpine lake at dawn with room on the left for a headline. 2k. Price?

Cursor sets the shape and size and calls:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "minimax-h3-image",
      "mode": "t2i",
      "prompt": "Ultra-wide hero banner: a calm alpine lake at first light, pine forest on the right edge, empty sky on the left half for a headline, crisp photographic detail",
      "resolution": "2k",
      "aspect_ratio": "21:9"
    }
  }
}

Cursor: One 2k image from MiniMax H3-Image is $0.036 (price list v15). Nothing is created by this estimate. Go ahead?

You: Go ahead.

{
  "name": "generate",
  "arguments": {
    "input": {
      "model": "minimax-h3-image",
      "mode": "t2i",
      "prompt": "Ultra-wide hero banner: a calm alpine lake at first light, pine forest on the right edge, empty sky on the left half for a headline, crisp photographic detail",
      "resolution": "2k",
      "aspect_ratio": "21:9"
    },
    "max_cost_usd": "0.036",
    "request_id": "alpine-lake-hero-21x9"
  }
}

When generate is called, Makernice quotes the input once more. A quote above max_cost_usd or the connection's per-job limit ends the call with a refusal and no job; otherwise the job is created and billed at that quote, and the response includes its job_id. If the agent repeats the call with the same request_id and input, it receives the existing job.

{
  "name": "check_job",
  "arguments": {
    "job_id": "gen_...",
    "wait_s": 45
  }
}

check_job returns once the image is done or the wait expires (at most 55 seconds). The image sits at outputs[0].url as a signed link that lasts 24 hours; call check_job again for a new link, or open console_url to find it in your console. An account webhook also covers jobs started over MCP.

The full set of text-to-image shapes is 1:1, 1:2, 2:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 9:21, 21:9. Leave resolution out and you get 1k. More on the model is on the MiniMax H3-Image model page.

How do I edit images with MiniMax H3-Image, and does it support layers or inpainting?

MiniMax H3-Image edits images in mode i2i: send 1 to 16 public https links in reference_images and write the change you want in prompt. Layers and inpainting are not supported, because the Makernice request has no mask or layer field, so describe the area to change in words instead.

References for MiniMax H3-Image are JPEG, PNG or WebP files of up to 10 MB. Edits have no aspect_ratio field; the result follows the reference images. The price is a base amount, $0.036 at 1k or $0.036 at 2k, plus $0.0096 for each reference image, the first one included. Edits with different numbers of references therefore get different quotes, so the agent should estimate after settling the list. MiniMax's guide sets each reference at 128 to 8192 px per side with a long-to-short ratio of at most 4:1, as listed on the image limits page; Makernice doesn't measure MiniMax H3-Image references before the job starts.

Here the agent prices a 1k group shot assembled from four portraits:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "minimax-h3-image",
      "mode": "i2i",
      "prompt": "Make a team photo: place the four people from the references side by side on the lake shore from the banner, same morning light",
      "reference_images": [
        "https://example.com/team/ana.jpg",
        "https://example.com/team/ben.jpg",
        "https://example.com/team/chen.jpg",
        "https://example.com/team/dora.jpg"
      ],
      "resolution": "1k"
    }
  }
}

Why did my MiniMax H3-Image generation fail, and is a failed task charged?

A failed MiniMax H3-Image generation is not charged: check_job shows status failed with an error object, billing moves to released and the hold is returned. Typical causes are a prompt or reference rejected by moderation, a reference link that redirects or can't be fetched, a reference outside the pixel or ratio limits, or a failure at the model provider. A finished image blocked by output moderation is still charged; the rules are on failed generations.

Refusals before a job exists are a separate case. Refusals reach the agent as readable text, and the call leaves no job and no charge. Column one is the name Makernice's server code uses for the refusal; column two, the beginning of the returned text.