# Can Claude Generate Images? What It Makes and How to Add It (2026) > Claude can't generate photos on its own. It draws SVG, charts and HTML visuals; connectors such as Adobe, Canva, Photoroom, fal or Kive add real image tools. Published: 2026-10-14 Updated: 2026-10-14 Author: Olga Stogova Tags: claude-image-generation, claude-mcp, model-context-protocol, ai-image-tools, ai-product-photography Kive studio: Driftwood Light Canonical: https://kive.ai/learn/can-claude-generate-images Claude cannot generate images in the sense most people mean. It produces no photos or illustrations itself; it reads images and writes text and code. What it can make natively are drawn visuals (SVG graphics, charts, diagrams and small HTML pages), and with an MCP connector added, it can drive an outside image tool that does produce real pictures. The photo above came from that second route: Claude called an image connector during the session that produced this article. ## Key takeaways - Anthropic's vision docs say Claude can interpret and analyze images but cannot generate, edit or manipulate them. - Natively, Claude builds SVG, HTML, Mermaid and chart visuals, and with file creation on it can save a chart as a PNG. - Custom connectors run on every Claude plan, but Free accounts are limited to one custom connector at a time. - Each connector bills in its own credits: Canva AI credits, Photoroom API credits, fal credits or Kive plan credits. - For product photos, pick a connector that starts from your product photo, not one that redraws the product from a prompt. --- ## What Claude makes on its own Claude draws with code. Anthropic's vision documentation puts it bluntly: Claude is an image understanding model only, and it cannot generate, produce, edit, manipulate or create images [[1]](https://platform.claude.com/docs/en/build-with-claude/vision). The Help Center adds the other half: it can build diagrams, charts and interactive visuals directly in a conversation, generated with HTML and SVG, a feature in beta on Claude web and desktop [[2]](https://support.claude.com/en/articles/9002504-can-claude-produce-images). Ask for "a photo of a serum bottle on driftwood" and Claude will tell you it doesn't produce photos. Ask for "an SVG illustration of a serum bottle" and it writes one – shapes, gradients and a label block, all as code you can edit or download. _An SVG Claude wrote for this article, exported to PNG_ | You ask Claude for | What happens natively | | --- | --- | | A photo or realistic illustration | Declines; it does not render raster images | | A logo sketch, icon or illustration as vector | Writes SVG code you can download | | A chart from your data | Builds it in chat, or saves a PNG with file creation on | | A flowchart or diagram | Writes Mermaid or SVG | | A small interactive page or tool | Builds an HTML or React artifact | | A change to your photo's background or color | Declines; it cannot edit pixels | With file creation enabled, Claude can save charts as image files like PNG and create artifacts such as SVGs and Mermaid diagrams [[3]](https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude). Those are rendered from code, not painted by a model, which is why they look like vector graphics no matter how carefully you prompt. ### What it reads Reading images is the stronger side. Claude accepts JPEG, PNG, GIF and WebP, up to 20 images per message on claude.ai, each up to 10 MB and 8000 x 8000 pixels [[1]](https://platform.claude.com/docs/en/build-with-claude/vision). Only the first frame of an animated GIF is read, and metadata is ignored, so Claude cannot tell you a photo's camera settings or whether a file carries an AI-generated tag. --- ## How connectors add image generation A connector gives Claude a tool it can call, and image tools are among them. Connectors use the Model Context Protocol, an open-source standard for connecting AI applications to external systems [[4]](https://modelcontextprotocol.io/docs/getting-started/intro). The image is made by the outside service; Claude writes the request, sends it and shows you what comes back. Custom connectors work on Claude, Cowork and Claude Desktop for Free, Pro, Max, Team and Enterprise plans, with Free users limited to one custom connector [[5]](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp). That limit is the first practical constraint: on Free you choose one image tool, not a toolbox. ### Add an image connector to Claude **Copy the tool's MCP server URL** Each vendor publishes one on its setup page. Photoroom's is `https://mcp.photoroom.com/mcp`; fal's is `https://mcp.fal.ai/mcp-relay`. **Open Customize, then Connectors** On Pro and Max, click + Add and choose Add custom connector, then paste the URL. On Team and Enterprise, an Owner adds it under Organization settings, then Connectors. **Sign in to the tool** The connector asks you to authorize your account with the vendor. That account's credits pay for every image. **Turn it on in the chat** Use the + button at the lower left of the chat, then Connectors, and switch the tool on for that conversation. In Claude Code the same server is one command, `claude mcp add --transport http` followed by a name and the server URL, and Anthropic recommends HTTP servers for remote connections ([Claude Code MCP docs](https://code.claude.com/docs/en/mcp)). Adobe, Canva and Figma are also listed in Claude's own connector directory. --- ## Image connectors for Claude, compared The connectors split into three jobs: editing a photo you already have, making a designed layout, and generating a new image. Each bills in its own credits, so the price of an image in Claude is the price of that tool. | Connector | What it does with images | Billed in | Setup | | --- | --- | --- | --- | | Adobe for creativity | Edits photos singly or in bulk: color, lighting, background removal, straightening, expand and crop | Adobe account; signing in raises usage limits | Claude connector directory | | Canva | Generates and edits designs, exports them | Canva AI credits; works on Canva Free | Claude connector directory | | Photoroom | Background removal, staging, shadows, relight, expand, upscale, Create Any Image | Photoroom API credits; $0.02 per cut-out on Basic, $0.10 per staged or edited image on Plus | Custom connector, mcp.photoroom.com/mcp | | fal | Runs fal's hosted image and video models | fal credits; OAuth sign-in, no API key | Custom connector, mcp.fal.ai/mcp-relay | | Kive | Product shots and video from a saved product, idea remixes, edits | Kive plan credits; 10 credits per remix image | Custom connector, mcp.kive.ai/mcp | Photoroom bills connector use at its API rates with no extra fee for Claude [[6]](https://www.photoroom.com/api/claude), and its API prices background removal at $0.02 per image on the Basic plan and AI edits such as product staging, backgrounds and shadows at $0.10 per image on Plus [[7]](https://www.photoroom.com/api/pricing). Adobe's connector lists 67 tools and edits photos individually or in bulk [[8]](https://claude.com/marketplace/connectors/adobe-creativity). Canva's connector is open to every Canva account, including Free, and generating designs may use your monthly Canva AI credits [[9]](https://www.canva.com/help/mcp-agent-setup/). fal connects through browser sign-in and charges generations to fal credits [[10]](https://fal.ai/docs/documentation/setting-up/mcp). Figma's connector is in the directory but is not an image generator: it reads design context and returns descriptions of screenshots rather than the images. The setup for Kive, from Kive's settings to a first product shot, is covered in [AI product shots with Claude MCP](/learn/claude-mcp-product-shots). _Generated by Claude through the Kive connector from a saved product – Kive studio: Warm Gradient_ --- ## Which route fits product photos For product photos the deciding question is where the product comes from. A prompt-only generator redraws the bottle from your description, so the label, cap and proportions are whatever the model guesses. A connector that starts from your photo keeps them. Photoroom works on the photo you upload: cut it out, relight it, stage it. Kive works from a saved product built from one to four photos of the item, and generates new scenes around it with a studio – the lighting, framing and setting saved as a reusable look. Both photos in this article came from that route: Claude called `generate_product_image` once per shot with a saved product and a studio, and each image used 10 credits from the workspace balance. Kive's [connect-to-Claude guide](https://kive.ai/docs/mcp/connect-to-claude) lists the setup. Kive is the wrong pick for one-off cut-outs on a white background, where a background tool is faster and cheaper per image, and for designed layouts with text, which is Canva's job. Whatever you use, compare each output with the real product before publishing; the [realism checklist](/learn/make-ai-product-photos-look-real) covers what to look for. ### Match the connector to the job Let Claude draw diagrams and charts itself, and add one connector for pictures. On the Free plan you get one custom connector, so choose by the job you repeat most. - **Charts, diagrams, icons:** Claude on its own – SVG, HTML and PNG charts need no connector - **Retouching photos you already have:** Adobe for creativity – color, lighting, background and crop edits, in bulk - **Social posts and layouts with text:** Canva – generates editable designs and works on Canva Free - **Cut-outs and staging at API prices:** Photoroom – $0.02 per cut-out on Basic; staging is $0.10 per image on Plus - **New product shots that keep your product:** Kive – generates from a saved product and a reusable studio - **Trying many open image models:** fal – one sign-in, many hosted models Claude's own drawing ability is better than its reputation: a clean SVG or a chart is often exactly what a deck or a docs page needs, and it costs nothing beyond your plan. The gap is pixels. If you test one connector first, run the same request through it twice and check whether the second image matches the first. Consistency across a set, more than the quality of any single picture, decides whether a tool belongs in your workflow. For the wider field of product-photo tools outside Claude, see the [AI product photography tools comparison](/learn/best-ai-tools-for-product-photography). ## Sources 1. [Claude Docs – Vision]() 2. [Anthropic Help Center – Can Claude produce images?]() 3. [Anthropic Help Center – Create and edit files with Claude]() 4. [Model Context Protocol – Introduction]() 5. [Anthropic Help Center – Custom connectors using remote MCP]() 6. [Photoroom – Edit images in Claude with Photoroom MCP]() 7. [Photoroom – API pricing]() 8. [Claude connectors – Adobe for creativity]() 9. [Canva Help – Canva AI Connector setup]() 10. [fal docs – MCP]() ## FAQ ### Can Claude generate images? Not photos or illustrations. Claude is an image understanding model: it reads images you upload but does not produce raster images itself. It can draw diagrams, charts and SVG or HTML visuals in chat, and it can call an image tool such as Adobe, Canva, Photoroom or fal once you add that tool as a connector. ### Can Claude edit images? Not by itself. Anthropic's vision documentation says Claude cannot edit or manipulate images. Add a connector to change pixels: the Adobe for creativity connector adjusts color and lighting, removes backgrounds and expands or crops photos, and Photoroom's connector removes backgrounds, relights, adds shadows and stages products, billed per edit. ### Is image generation in Claude free? Claude does not charge extra for a connector, but the image tool behind it bills in its own currency, such as Canva AI credits, Photoroom API credits or fal credits. Custom connectors work on every Claude plan, though Free users are limited to one custom connector at a time. ### What image files can Claude read? Claude reads JPEG, PNG, GIF and WebP. On claude.ai you can attach up to 20 images per message, each up to 10 MB and 8000 x 8000 pixels. Only the first frame of an animated GIF is used, and Claude does not receive the image's metadata. ### How do I add an image generator to Claude? On Pro or Max, open Customize, then Connectors, click + Add, choose Add custom connector and paste the tool's MCP server URL, then sign in to that tool. Team and Enterprise owners add it under Organization settings. In Claude Code, run claude mcp add with the http transport and the same URL. ### Which Claude connector keeps my real product in product photos? Use one that works from your product photo rather than a text prompt. Photoroom stages and relights the photo you upload. Kive's connector generates new shots from a saved product built from your photos, so labels and shape carry over, and it charges from the same credits as the Kive app.