Short answer: Claude has no built-in image generator. Anthropic's help center says Claude does not generate photos or illustrations; it draws charts and diagrams in HTML and SVG and analyzes images you upload. To get real images out of Claude in 2026 you connect a tool: an MCP server (remote or local), a Claude Skill that calls an image API, or a connector from an image platform. This guide compares every working route and shows the three-step setup.
Updated November 2026. Avocado AI publishes this blog and runs one of the MCP servers compared below. Prices are in euros where they are ours; competitor facts come from the linked pages.
Does Claude have image generation built in?
No. The official answer is in Anthropic's help article Can Claude produce images?: "Claude doesn't generate photos or illustrations the way image-generation tools do." What Claude can do natively is build diagrams, charts and interactive visuals in the conversation, rendered from HTML and SVG (a beta feature on web and desktop), and view and analyze images you upload.
That is the whole native story, and it has not changed with Claude Pro or Max. A paid Claude plan gives you more usage, longer context and features like Projects, not a pixel renderer. ChatGPT and Gemini ship an image model inside the chat; Claude does not.
So "claude image generation" in practice means "Claude plus a tool that generates images". The good news is that Claude is built for tool calling, and the Model Context Protocol (MCP), an open standard Anthropic created, makes adding that tool a two-minute job.
Every way Claude can generate images in 2026
There are five routes. They differ in where the model runs, who pays for it, and where the finished image ends up.
| Route | How it works | Setup effort | Cost model | Where images land | Video too? |
|---|---|---|---|---|---|
| Native Claude | Charts, diagrams, SVG and HTML visuals only | None | Included in your Claude plan | In the chat, downloadable as .svg or .html | No |
| Claude Skill that calls an image API | A SKILL.md folder with a script that hits OpenAI, Gemini or another image API with your own key | Medium: install the skill, enable code execution, add a key | Pay per image to the API provider | A file in the sandbox or your repo | Rarely |
| Open-source local MCP server (GitHub) | A Node or Python process on your machine wraps one provider's API | Medium to high: Node or uv, config file, API key, restarts | Pay per image to the API provider | A folder on your disk | A few do |
| Image platform connector (Higgsfield, OpenArt) | A remote MCP server you add as a custom connector; sign in with the platform account | Low | The platform's credits | The platform's library | Yes |
| Workspace MCP server (Avocado AI) | A remote MCP server with 60+ tools; outputs are saved to a brand workspace with product references and rules | Low | Avocado credits, cost shown before each run | Your Avocado workspace and gallery | Yes |
A note on the middle rows. The GitHub servers are real and useful if you are a developer with an API key already. Our roundup of image generation MCP servers goes through nine of them with facts from their READMEs. The short version: most are one-provider wrappers with a handful of stars, they run as a local process, and the image lands in a folder, not in a place your team can find later.
Route 1: Claude's native visuals
Use this when the output is a chart, a flow diagram, a wireframe or an icon. Upload a CSV, ask "what does the data show", and Claude renders an interactive chart you can download. It will not give you a product photo, a lifestyle image or an ad creative, and asking harder does not change that.
Route 2: Claude Skills
Per Anthropic's What are skills? article, skills are "folders of instructions, scripts, and resources that Claude loads dynamically." They need code execution enabled and are available on Free, Pro, Max, Team and Enterprise plans. Anthropic's own skills cover Excel, Word, PowerPoint and PDF files; none generates images. The community has published image-gen skills that call OpenAI or Gemini image APIs with your key. They work, but you are installing someone else's script, so read it first, and the output is a file, not a managed asset.
Anthropic itself draws the line this way: MCP connects Claude to external services and tools, skills supply procedural knowledge about how to use them. For image generation, the tool is the hard part, which is why MCP is the main route.
Route 3: open-source MCP servers from GitHub
Projects like nanobanana-mcp-server (Gemini image models, 404 stars at the time of writing, runs with uvx and a GEMINI_API_KEY) or openai-gpt-image-mcp (gpt-image-1, 119 stars, needs an OPENAI_API_KEY from a verified organization) give you one provider's image model inside Claude Desktop or Claude Code. You edit a JSON config, restart the app, and you are generating.
Trade-offs: one model family per server, your own API billing, local-only (the Claude web app cannot reach a process on your laptop), and no brand memory. Fine for a developer making icons; awkward for a marketing team that needs the same product to look right across fifty images.
Route 4: image platform connectors
Higgsfield's Claude image generator page describes an MCP connector you add in Claude's connector settings, with "15+" models named on the page (Nano Banana Pro, Soul, Seedream, FLUX, GPT Image, Kling O1, Wan), free renders to start and credits after that. OpenArt's MCP page lists eight image and eight video models behind https://mcp.openart.ai/mcp, OAuth sign-in and the same credits as its web app. Both are remote, so they work in Claude on the web, desktop and mobile. Outputs save to that platform's library.
Route 5: a workspace MCP server (what we built)
The Avocado MCP server is also a remote connector, so setup is the same two minutes. The difference is what sits behind the tools. Every image, video, voiceover and music track is saved to your Avocado workspace, Claude can read your Brand DNA (products, references, colors, rules) before it prompts, and the credit cost is shown before each generation. It is included on every paid plan from Intro at 19.99 EUR a month for 100 credits. There is no free plan and it is not a local server, which matters if you want to run everything offline.
How to generate your first image in Claude (3 steps)
This uses the Avocado connector. The first two steps are identical for any remote MCP server, and we wrote a separate, vendor-neutral guide on how to add an MCP server to Claude.
Step 1. Add the connector. In Claude on the web or desktop, open Settings, then Connectors, choose Add custom connector and paste https://mcp.avocadoai.co/api/mcp. Claude opens the Avocado sign-in; approve it. In Claude Code, run claude mcp add --transport http avocado-ai https://mcp.avocadoai.co/api/mcp --header "Authorization: Bearer YOUR_KEY" with a key from your Avocado API keys page.
Step 2. Start a new chat and enable it. Tool lists load when a conversation starts, so open a fresh chat. Click the + button at the lower left, choose Connectors and make sure the toggle is on.
Step 3. Brief it like a designer. Try: "Generate a 4:5 Instagram ad for our vitamin C serum on a coral background, soft morning light." Claude calls generate_image, Avocado picks the model (GPT Image 2.5 by default, Nano Banana 2 when you point at a saved product or Element), and the finished image is saved to your workspace with full commercial rights. Follow up with "swap the background for a bathroom shelf" and Claude calls edit_image on the same asset.
If you want to pick the model yourself, name it: "use Nano Banana 2" or "use Seedream 5.0 Pro". The models page lists what is live, and the Nano Banana 2 and GPT Image 2.5 pages explain when to pick which.
Which image model should Claude use?
Claude does not care which model is behind the tool; you do. Here is the split we see across ecommerce and DTC customers.
| Job | Model to ask for | Why |
|---|---|---|
| Product on a clean background, sharp label text | GPT Image 2.5 | Best text rendering, up to 4K |
| Fast variations of a saved product photo | Nano Banana 2 | Reference-driven, quick, keeps the product consistent |
| Editorial lifestyle scenes | Seedream 5.0 Pro | Strong on skin, fabric and light |
| Stylized or illustrated ads | Recraft v4.1, Ideogram v3 | Vector-like output, typography |
| Mood boards and style exploration | Flux 2, Krea v2 | High prompt adherence, fast iterations |
All of these are available by name through the Avocado connector and in the AI image generator in the web app.
Can Claude generate video?
Same answer, same mechanism. Claude has no built-in video model, and with an MCP server it can call one. Through the Avocado connector, generate_video exposes Seedance 2.5 (up to 30 seconds), Kling 3.0 (up to 4K), Veo 3.1 and MiniMax H3. Video jobs run in the background: the server returns a job id right away, Claude keeps working, and it polls check_job until the clip is in your workspace. Expect two to five minutes depending on duration and resolution.
A working flow from a chat: "Make a product photo of the serum on a wet stone slab", then "turn that into a 5 second Seedance 2.5 video with a slow push-in", then "write a 15 second voiceover and make a 9:16 version for TikTok". Each line is one tool call, and the assets stay linked in your workspace. For the full video-side comparison, see Claude video generation for ecommerce brands.
Common mistakes when generating images with Claude
- Asking Claude to "draw" without a tool connected. You get an SVG or a polite refusal. Connect a server first.
- Expecting Claude Pro to unlock images. It does not. Image generation comes from the tool, not the plan.
- Running a local server and wondering why claude.ai cannot see it. Local stdio servers only work in Claude Desktop and Claude Code. The web app only reaches remote servers.
- Vague briefs. "Nice product photo" gives you a stock-looking render. Name the product, the surface, the light, the aspect ratio and the platform.
- Forgetting the new chat. After adding a connector, tools appear in conversations started afterwards.
FAQ
Does Claude Pro generate images?
No. Claude Pro, Max, Team and Enterprise all share the same native limits: charts, diagrams and SVG visuals, plus image analysis. Photo-style generation needs a connected tool on every plan. Custom connectors are available on Free, Pro, Max, Team and Enterprise, with Free limited to one connector, per Anthropic's connector article.
Can Claude Code generate images?
Not by itself, but Claude Code is the easiest Claude surface to extend. One command, claude mcp add --transport http name url --header "Authorization: Bearer key", adds a remote server, and local stdio servers are added with claude mcp add name -- npx package. After that, "generate a hero image for this landing page" is a normal instruction.
Is Claude image generation free?
Claude's native visuals are included in every Claude plan. Generated photos and ad images are billed by whichever tool you connect: an API key for the GitHub servers and skills, platform credits for Higgsfield and OpenArt, and Avocado credits for the Avocado connector, which starts at 19.99 EUR a month for 100 credits with no free plan.
Which model does Claude use for images?
Whichever one the tool exposes. Single-provider servers give you one family (Gemini, gpt-image). Platform and workspace connectors let Claude pick or let you name the model: GPT Image 2.5, Nano Banana 2, Seedream 5.0 Pro, Flux 2 and others through Avocado.
Can Claude edit an existing image?
Natively it can only describe or analyze an uploaded image. With an image tool connected, edit tools accept your image and a prompt: background swaps, label fixes, new angles. Through Avocado, edits default to GPT Image 2.5 and build on the asset you already approved.
Does Claude save the images it generates?
Natively, visuals stay in the chat until you download them. With a local MCP server, images go to a folder on your disk. With a platform or workspace connector, they are saved to that account's library, which is the only option where a teammate can pick the asset up later.
Next step
If you only need a chart, open Claude and ask. If you need product images and video that match your brand, connect the Avocado MCP server, start a new chat, and brief Claude like you would brief a designer. Plans start at Intro, 19.99 EUR a month, and every paid plan includes the connector, full commercial rights and no watermark.