Does Claude have a built-in image generator?

No. Anthropic's help article says Claude doesn't generate photos or illustrations the way image-generation tools do. It can make diagrams, charts and interactive visuals with HTML and SVG, in beta on web and desktop, and it can look at images you upload.

Can Claude create images for free?

Claude can draw SVG and HTML visuals in chat. For pictures from a model, a custom connector works on the Free plan, which allows one, but the image model you connect usually charges for its images.

Can Claude Code generate images?

Yes, in two ways. It can write code that renders an image on your computer, as our OG Image and YouTube Thumbnail skills do, at no cost per image. Or it can call an image model through a connector or an API, and you pay for each image. Claude Code needs a paid Claude plan.

Can Claude edit a photo I upload?

Claude can look at your photo and describe it, but changing it into a new realistic picture needs an image model. Glidezo's online skills do that for a few set jobs, such as Virtual Staging and Day to Twilight.

Why does the SVG look so simple?

Because it is shapes written by hand as code, not a picture made by a model. SVG is good for logos, icons and diagrams. For anything that should look like a photograph, use an image model.

Can Claude Generate Images? What It Can and Can't Do

Updated 2026-11-02

Not the way an image generator does. Anthropic says Claude doesn't generate photos or illustrations the way image-generation tools do. It can draw with code, such as SVG and HTML, and write code that renders a finished PNG. For a realistic picture, it needs an image model through a connector.

A weathered red wooden rowing boat on a still, misty lake at dawn, with pine trees on the far shore and a pink sky. — From an image model, Nano Banana 2.1, not from Claude. A real run of our Film Look Image skill, not retouched. It cost us $0.0336. Scene: "A small red wooden boat on a still lake at dawn, with pine trees on the far shore".

A flat illustration drawn with code of a small red boat on a pink lake at dawn, with dark mountains and pine trees behind it. — From Claude, with no image model. Claude Code wrote this as about 25 lines of SVG code for the same scene, with one small fix to a mountain edge. It cost nothing beyond the Claude plan. Shapes, not a photo.

From code. Our free OG Image skill laid this card out as a web page and saved it as a 1200x630 PNG on our computer. The text is exact, which is what image models often get wrong.

All three pictures above are real and we made them ourselves. None is retouched, and the one fix to the SVG is in its caption. Glidezo isn’t connected to Anthropic. We sell credits for image runs, so weigh what we say about that route with that in mind.

The three routes at a glance

Question	Claude drawing with code	Code rendered to a file	An image model through a connector
Good for	Logos, icons, diagrams, charts	Cards and thumbnails with exact text	Photos, illustrations, anything realistic
Bad at	Anything that should look real	Anything that should look real, unless you supply the photo	Exact text, the same result twice
Needs	Any Claude plan	Claude Code (paid plan), Node.js, FFmpeg	Any Claude plan, plus an account with the image provider
Cost per image	Nothing extra	Nothing extra	The model’s price. Ours cost $0.0336

Route 1: Claude draws with SVG and HTML

Ask Claude for a logo, a diagram or a chart, and it writes the drawing as code. In the Claude app these visuals are in beta on web and desktop, and you can ask for changes in chat.

Here are a few lines of the code behind the second picture above, and you can see why it looks like shapes:

<rect width="1376" height="430" fill="url(#sky)"/>
<circle cx="930" cy="372" r="58" fill="#fff0c9"/>
<rect y="430" width="1376" height="338" fill="url(#water)"/>
<g transform="translate(470 500)">
  <path d="M0 40h340l-46 68H44z" fill="#b32d2d"/>
</g>

What it can’t do. It can’t make a photograph or a detailed illustration. Faces and hands drawn this way won’t look real.

Route 2: code rendered to a PNG

Claude Code can write a web page and save it as an image, with exact text and an exact size. Install our free skills, then ask for what you need:

npx skills add glidezo/skills

OG Image makes link preview cards and YouTube Thumbnail makes 1280x720 thumbnails. Both render with HyperFrames on your computer, so there is no cost per image.

What it can’t do. It lays out what you give it: your text, your logo, your screenshot. It won’t invent a new picture. Claude Code also needs a paid Claude plan.

Route 3: connect Claude to an image model

A connector lets Claude call another service, so Claude can write the prompt and an image model makes the picture. Custom connectors work on every Claude plan, and Free allows one. For example, in Claude Code, using Sume, the provider Glidezo’s own runs use:

claude mcp add --transport http sume https://mcp.sume.com/mcp
claude mcp login sume

Sume’s docs say an image takes about 10 to 60 seconds. We didn’t connect Claude this way for this page: the boat above came from our own online skill, which calls the same provider directly. Other connectors exist, and we haven’t checked them. For prices, see Nano Banana pricing.

What it can’t do. You pay for every image, even the ones you throw away, and image models can still garble text.

Which should you pick?

A logo, icon or diagram: ask Claude for SVG.
A card or thumbnail with words that must be right: use code.
A photo-like picture: use an image model.

Or skip Claude: run one here

If you just want the picture, these skills run online with no install, and the price shows before you run them. The boat above took 5 credits (5 cents) on Film Look Image.

Film Look Image: a scene in a film look, from your words
Sketch to Realistic Image: a drawing into a realistic picture
Virtual Staging: furniture in an empty room photo

What these can’t do. Each one does a set job, not anything you can describe. Results can have mistakes: our sketch sample changed some doors and walls. People’s photos have extra rules, in our content rules.

Want video instead? See can Claude make videos? For more free projects you can run yourself, see open source.

Questions

Does Claude have a built-in image generator?: No. Anthropic's help article says Claude doesn't generate photos or illustrations the way image-generation tools do. It can make diagrams, charts and interactive visuals with HTML and SVG, in beta on web and desktop, and it can look at images you upload.
Can Claude create images for free?: Claude can draw SVG and HTML visuals in chat. For pictures from a model, a custom connector works on the Free plan, which allows one, but the image model you connect usually charges for its images.
Can Claude Code generate images?: Yes, in two ways. It can write code that renders an image on your computer, as our OG Image and YouTube Thumbnail skills do, at no cost per image. Or it can call an image model through a connector or an API, and you pay for each image. Claude Code needs a paid Claude plan.
Can Claude edit a photo I upload?: Claude can look at your photo and describe it, but changing it into a new realistic picture needs an image model. Glidezo's online skills do that for a few set jobs, such as Virtual Staging and Day to Twilight.
Why does the SVG look so simple?: Because it is shapes written by hand as code, not a picture made by a model. SVG is good for logos, icons and diagrams. For anything that should look like a photograph, use an image model.

Sources

Anthropic Help: Can Claude produce images? (article dated 2026-03-16) (checked 2026-11-02)
Anthropic Help: custom connectors (every plan, one on Free) (checked 2026-11-02)
Claude pricing: Claude Code is on Pro and above, not Free (checked 2026-11-02)
Sume docs: MCP quickstart (images take about 10 to 60 seconds) (checked 2026-11-02)