Can Claude create product images? From your photo, via a tool

Not by itself: Claude can look at a product photo, but it cannot create images. It can still get you product images by calling an image tool: Claude sends your real product photo as a reference image, describes the new scene or background in the prompt, and the image model renders it.

Claude's side comes from Anthropic's vision docs and custom connectors article. The tool side comes from Sume's Image API and MCP tools and gates docs and its current MCP tool descriptions. All were read on 2026-09-29.

What can Claude do with a product photo on its own?

Read it. Anthropic's vision docs say Claude can interpret and analyze images but cannot generate, produce, edit, manipulate, or create them. So Claude can describe your product, draft scene ideas, and write the prompts; the pixels have to come from an image tool it calls.

How do I make product images from Claude?

Connect an image tool as a custom connector. Anthropic says custom connectors using remote MCP work on Claude, Cowork, and Claude Desktop on every plan, with Free limited to one. Sume has no official Claude connector: you add its hosted MCP server, https://mcp.sume.com/mcp, and turn Write on at Sume's consent page. Sume's basics page says hosted MCP still works but is not the primary path today. Then:

• Put the product photo at a public HTTPS URL, such as the image on your store page. Hosted MCP cannot read files from your laptop, and localhost, private-network, and non-HTTPS URLs are rejected.
• Ask Claude for one scene per request: the setting, the light, and the aspect ratio. Claude calls generate_image with your photo in input_references.
• Pass aspect_ratio: "auto" to keep the reference's shape; omitting the field is not the same as auto. For a different shape, name a ratio the model lists, such as 1:1 or 4:5.
• Check every result against the real product before you publish it.

{
  "idempotency_key": "product-scene-001",
  "dry_run": true,
  "payload": {
    "model": "openai/gpt-image-2.5",
    "prompt": "The bottle on a marble counter, soft morning light",
    "aspect_ratio": "auto",
    "input_references": [
      {
        "type": "image_url",
        "image_url": { "url": "https://example.com/bottle.jpg" }
      }
    ]
  }
}

Which models take a product photo as a reference?

Not all of them. Sume's docs say models whose input_references descriptor is {"min": 0, "max": 0} are text-to-image only and reject references, so Claude should read the catalog with image-models_list before picking one.

Can Claude make marketing images or a white background?

Yes, the same way. A marketing image is a product scene with room for your copy; ask for the empty space in the prompt and add the text in a layout tool. For a white background, describe it in the prompt with your photo as the reference. For a transparent one, Claude can cut the product out with rmbg_create, which returns a PNG with alpha; AI image generator with a transparent background covers that step. AI lifestyle product photography covers prompts for lifestyle scenes.

How much does it cost?

Each image is billed to your Sume wallet, not to your Claude plan, and the price depends on the model. Send the call with dry_run=true first, as in the example: it previews the cost without submitting the job. Then send it again without dry_run to render. Image billing is all-or-nothing: a completed image is billed in full and a failed one is not billed.