Can Claude Generate Images? Claude Image Generation Through MCP

No. Claude cannot generate photos, illustrations or video on its own. Anthropic’s vision documentation calls Claude “an image understanding model only”: it reads the images you give it and answers in text. What Claude can do, as of October 2026, is run a separate image or video generator through the Model Context Protocol (MCP). Claude writes the prompts and sends the batches, and the generator makes the pixels.

What Claude can and cannot do with images and video

What you ask Claude for	Native?	What you get
A photo or illustration	No	Nothing, unless a connected tool makes it
A video clip	No	Nothing, unless a connected tool makes it
Changes to what is in a photo	No	A text description of the changes
A crop or resize of an image file	Yes, with code execution on	A new file from code Claude runs
A read of an uploaded image	Yes	A text answer
A diagram, chart or interactive visual	Yes, as code	HTML and SVG drawn in the chat
An SVG image or flowchart	Yes, as code	An artifact you can edit and reuse
Generated images or video through MCP	Through a connected tool	The tool’s output, with Claude running the requests

What Claude can do with images on its own

Claude works with images in two directions. It reads them well, and it can draw with code. It cannot paint pixels.

It reads images

You can upload an image on claude.ai or drag it into the chat, and developers can send images through the API. Anthropic’s models overview states it plainly: “All current models support text and image input, text output.”

For a creator, this is more useful than it sounds. You can upload a finished picture and ask Claude whether the lighting matches the rest of a set, or whether the outfit matches the brief you wrote.

It draws with code

Anthropic’s help article Can Claude produce images? opens with the limit: “Claude doesn’t generate photos or illustrations the way image-generation tools do.” The same article says Claude can “build diagrams, charts, and interactive visuals directly in your conversation,” generated with HTML and SVG. That feature is in beta on Claude web and desktop.

Artifacts are the longer-lived version. Anthropic lists SVG images, diagrams and flowcharts, single-page HTML websites and interactive React components among the common artifact types. Artifacts need “Code execution and file creation” turned on in Settings, under Capabilities (menu paths in this guide are as of October 2026).

These visuals are real and useful. A chart of last month’s sales by content type, or a flowchart of who on your team approves which set, are both quick to get. None of them is a photo.

What it cannot do

The vision documentation is specific: Claude “can interpret and analyze images, but it cannot generate, produce, edit, manipulate, or create images.”

So Claude alone cannot:

• Create a photorealistic person, real or AI.
• Change what is in a photo: retouch skin, relight a scene, swap an outfit or replace a background.
• Upscale an image with real added detail.
• Keep a face consistent across a set, because it produces no faces at all.

There is one narrow exception. With code execution turned on, Claude can write and run code that crops, resizes or converts an image file. That changes the file, not what is in the picture.

If you came here asking “can Claude make images” for a content library, the honest answer is that Claude can plan the library but not shoot it.

Can Claude generate videos?

No. Claude video generation does not exist as a native feature. Every current model gives text output, so no video file comes out of Claude itself.

The route is the same as for images. Connect Claude to a video generator through MCP, and Claude can describe the shot, choose which still image to animate, and request several variations. The video model does the rendering. If you want to know what one of those video models handles well, our Seedance 2.0 guide covers it.

How MCP lets Claude call an image or video generator

MCP (Model Context Protocol) is “an open-source standard for connecting AI applications to external systems,” according to its official documentation. Anthropic’s connectors overview notes that Anthropic created it. A service publishes an MCP server, the server exposes tools, and Claude can call those tools during a conversation.

For image and video work, that splits the job cleanly:

• Claude does the language work. It turns your brief into prompts, splits a set into batches and keeps track of what came back. Where the files end up depends on which Claude you use, as covered below.
• The generator does the image work. It runs the image or video model and returns the files. Whether faces stay consistent depends on the generator.

Claude never makes the pixels in this setup. The output comes from the generator’s models, under the generator’s rules and on the generator’s plan.

Where you connect it

Anthropic documents three ways to add an MCP server, depending on which Claude you use.

Where you use Claude	How Anthropic documents the connection
Claude on the web, desktop and mobile	A custom connector using remote MCP: Customize, then Connectors, then add a custom connector with the server URL and sign in
Claude Desktop with a local server	Desktop extensions under Settings, then Extensions, or the local configuration file
Claude Code	claude mcp add with the server’s transport and URL, or a local server; check the status with /mcp

A few details from Anthropic’s custom connector guide matter in practice:

• Custom connectors are available on Claude’s Free, Pro, Max, Team and Enterprise plans. Claude’s free plan is limited to one custom connector.
• On Claude’s Team and Enterprise plans, an Owner adds the connector to the organization first. Members then connect to it individually.
• You switch connectors on or off per conversation with the “+” button in the chat, under Connectors.
• Claude reaches a remote server from Anthropic’s cloud, not from your computer, so the server has to be reachable over the public internet.

Claude Desktop’s local MCP servers are a separate mechanism that runs on your machine. Claude Code recommends HTTP servers for remote services and also supports local servers. Because Claude Code also runs commands in your terminal, it can work with a tool’s command-line interface as well as its MCP server.

The surface also decides what happens to the files. On claude.ai with a remote connector, Claude cannot write to folders on your computer; it shows the results the tool returns in the chat. Claude Code, or Claude Desktop with a local filesystem server, can save files into a folder you choose.

The same custom connector guide ends with security advice worth repeating. Only connect to servers from organizations you trust. Review the permissions a server asks for, and watch for prompt injection from malicious servers.

Where FansCreator fits

FansCreator generates realistic images and video with one consistent face and body, for creators and agencies on OnlyFans, Fanvue, MYM and Fansly. Check each platform’s AI rules below before posting. Besides the web app, it offers an API, an MCP server and a CLI, plus agent skills. These expose the product’s features to any AI assistant: image and video generation, face swap, character swap, editing and upscaling.

The MCP server and CLI are included on FansCreator’s Basic, Pro and Enterprise plans, not on Starter. The connection details, such as the server address and how you sign in, come from your FansCreator account. Add them through the Anthropic flow above that matches your Claude surface. FansCreator does not publish a public setup page, so ignore setup instructions from third-party sites.

The split of work stays the same as with any MCP generator:

• Claude plans the set, writes the prompts, runs the batches and shows the results the FansCreator tool returns.
• FansCreator runs the generation with models such as Wan, Seedance, Seedream, Qwen-Image and Z-Image. It keeps the creator’s face and body the same across pictures and clips. How that works for video is covered in how to keep one face across AI video clips.

FansCreator does not post anything. You download the files and publish them yourself.

Example: one persona in 10 outfits for an Instagram teaser set

Here is a typical SFW job where Claude saves time.

1. Create the persona in FansCreator first. Design her from scratch with the AI influencer generator, choosing hair, eyes, body, ethnicity and skin tone, or start from your own reference photos. Claude cannot design a face. The persona lives in FansCreator.
2. Give Claude the brief. For example: “Using my FansCreator persona, plan 10 autumn outfits for an Instagram teaser set. One city location per outfit, natural daylight, waist-up framing. Write one prompt per image and show me the list before you generate anything.”
3. Edit the list. Swapping an outfit in a text list costs nothing. Swapping it after generation costs another render.
4. Let Claude run it. Ask it to generate in two batches of five. In Claude Code, or Claude Desktop with a local filesystem server, you can also have it save the results into one folder, named by outfit and location. On claude.ai, it shows the results the FansCreator tool returns in the chat.
5. Review the results. Look at every picture yourself. You can also upload a picture to Claude and ask whether it matches the brief.
6. Post them yourself. Pick the best images and publish them on Instagram on your own schedule.

The value is in the repetitive parts: ten prompts that stay consistent in style and batches you don’t have to babysit. For a single picture, the FansCreator app is faster than a conversation.

Claude is not the route for NSFW content

Anthropic’s Usage Policy has a section titled “Do Not Generate Sexually Explicit Content.” It prohibits using Anthropic’s products to “Depict or request sexual intercourse or sex acts.” It also rules out using them to “Generate content related to sexual fetishes or fantasies” or to “Engage in erotic chats.” The policy also states that “Agentic use cases must still comply with the Usage Policy.”

Connecting a generator does not change this. The request still passes through Claude, so Claude’s rules apply to it. Keep Claude workflows to SFW work: teasers, Instagram and TikTok sets, X previews and other promo material.

Adult creators who need explicit content should make it directly in a tool built for it. FansCreator’s NSFW AI image generator works in the FansCreator app itself, not through Claude.

Consent applies either way. Only use your own face, your own AI persona, or a face whose owner gave you provable consent. FansCreator’s Terms of Use prohibit anything else.

What each platform allows for AI content

Each platform sets its own AI rules, and they differ more than most creators expect:

• OnlyFans. AI content is allowed only if it “clearly features the verified Creator” and fans can tell it is AI, for example with “#ai”, per the OnlyFans Help Centre and the Terms of Use. The Help Centre also lists content that is wholly AI-generated, or that “features heavily altered images”, among content it removes. A persona built from scratch does not belong on OnlyFans.
• Fanvue. Fully AI-generated creators are allowed with a “clear and prominent” disclosure in the bio, the caption or a watermark, per Fanvue’s AI content article. Face-swapped content is allowed if you verify the real model behind it.
• MYM. AI content is allowed but requires a “Virtual Image”, #AI or #AIgenerated label, per MYM’s AI content guide.
• Fansly. “Photorealistic AI-generated content is not allowed,” and labelling it as AI does not make it permissible, per Fansly’s AI content article. Its terms name face-swapped content too. Only non-photorealistic virtual personas can apply.

Getting started with Claude and FansCreator

If you only need diagrams, charts or a second opinion on your pictures, Claude already does that with no setup.

If you want Claude to produce images and video, you need a generator behind it. To use FansCreator, create an account and start on the free plan. Connecting Claude through MCP or the CLI requires FansCreator’s Basic, Pro or Enterprise plan. From there, Claude handles the prompts and FansCreator makes the pictures.