Black Friday starts early: Save 50% on Pro.Explore plans
Guide

Can Claude generate images?

Yes. Claude Code can already write code that renders images, from SVG and charts to OG cards. Add an image Skill and the style, layout and prompts come pre-designed, so the results match what you had in mind.

Image Skills to start with

Claude Style Illustration$4.99

Paste into your agent

Read https://skillry.dev/skills/bs-claude-style-illustration/install.md and follow the instructions to install the skill.

Linktree Style OG Image$4.99

Paste into your agent

Read https://skillry.dev/skills/bs-linktree-style-og/install.md and follow the instructions to install the skill.

Headspace Style OG ImageFree

Paste into your agent

Read https://skillry.dev/skills/bs-headspace-style-og/install.md and follow the instructions to install the skill.
Short answer

Yes. In Claude Code, Claude can write code that draws an image (SVG, a chart script or an HTML page rendered to PNG), but you describe every colour, font and layout yourself. An image Skill brings those already designed. For illustrations, posters and photo-style pictures it writes the prompt and calls an image model through your fal API key. In the claude.ai chat, Claude doesn’t hand you an image file.

How to generate images with Claude, step by step

Claude writes the prompt and drives the Skill; fal renders the image.

  1. Hand selecting one of three wireframe Skill cards
    Step 1

    Choose an image Skill for the output

    Pick the Skill built for what you will publish: an OG image, poster, infographic or illustration.

    Each image Skill on Skillry is built for one job, such as a 1200×630 OG image, an event poster or an infographic, and carries its own layout, type and colour rules. Open a Skill page to see real examples first.

    See all image Skills
  2. Hand sliding a Skill card into an open folder
    Step 2

    Install the image Skill in Claude Code

    Copy the install prompt from the Skill page and paste it into Claude Code.

    Read https://skillry.dev/skills/<slug>/install.md and follow the instructions to install the skill.

    The Skill lands as a SKILL.md folder: ~/.claude/skills/ for Claude Code, ~/.agents/skills/ for Codex, and the matching folder for Cursor, Antigravity, OpenCode or GitHub Copilot.

  3. Hand holding a key toward a lock, representing a fal API key
    Step 3

    Add your fal API key

    Create a key in your fal account and set it as FAL_KEY.

    export FAL_KEY=<your-fal-key>

    Create an API key in your fal account and set it as an environment variable named FAL_KEY in the shell or project where your agent runs. The Skill reads that variable when it calls fal’s image models. Keep the key out of your repository: a git-ignored .env file or your shell profile both work.

  4. Speech bubble connected to a picture frame
    Step 4

    Ask Claude in plain words, then refine

    Describe the image; Claude writes the prompt, calls fal and saves the file.

    /<slug> “Launch post: v2 is live”

    Describe the topic, audience, headline and where the image will be used. If something is off, ask for a different crop, palette or headline and Claude re-prompts with the Skill’s style rules.

Claude alone vs Claude + fal

AspectClaude aloneClaude + image Skill + fal
Reads and describes imagesYesYes
Outputs PNG / JPG filesNoYes
Same style across imagesNoYes
Preset OG, social and print sizesNoYes

Same Skill, same fal key, any agent

Notion Style OG Image Skill page with logos of Claude Code, Codex, Cursor, Antigravity, OpenCode and GitHub Copilot

One setup for every agent

Image Skills are plain SKILL.md folders, and fal works the same everywhere.

  • Claude CodeClaude Code
  • CodexCodex
  • CursorCursor
  • AntigravityAntigravity
  • OpenCodeOpenCode
  • GitHub CopilotGitHub Copilot

What Claude image generation can and can’t do

What Claude does on its own

Claude reads images, describes them, critiques layouts and writes code that draws: SVG icons, HTML pages, chart scripts and diagrams. In Claude Code it can render that code to a PNG. What it does not do on its own is paint a new illustration or photo-style picture the way an image model does. Ask Claude for a picture in chat and you get a description, an SVG or a diagram. That is a limit of the output format, not of Claude’s visual judgment.

fal docs page Get Your API Key next to a FAL_KEY environment variable card

falHow fal fills the gap

fal hosts image models behind one API. A Skillry image Skill tells Claude how to turn your request into a precise prompt, which size to use and which style rules to follow, then calls fal with your FAL_KEY and saves the result into your project. Claude stays in charge of the brief and the prompt; fal does the rendering. You pay fal directly for what you generate.

Other ways: MCP image servers

Claude can also call an image model through an MCP server, such as a Hugging Face Space connected to Claude. That suits one-off pictures, but you choose the model, prompt and size every time. Image Skills are the repeatable route: style, sizes and prompt rules are saved once, so every image matches.

More on Claude image generation

Claude image generation compared with ChatGPT and Codex.

Claude image generation vs ChatGPT and Codex

ChatGPT and Codex can generate images with their built-in GPT Image 2. Claude Code, Cursor, Antigravity, OpenCode and Copilot have no built-in image generation, so they use fal through a Skill. Skillry image Skills use fal by default in every agent, so the same Skill, prompts and FAL_KEY give matching results whichever agent your team uses.

Claude image generation FAQ

Can Claude generate images directly?

Partly. In Claude Code, Claude writes SVG, HTML or chart code and can render it to PNG, which covers diagrams, charts and text-led cards. Painted illustrations, posters and photo-style images need an image model: an image Skill writes the prompt, calls fal with your key and keeps every image in one designed style. Claude can also read and critique images you give it.

Does Claude generate images in the Claude app?

No. In the Claude app, Claude doesn’t create photos or illustrations the way image tools do, but it can make diagrams, charts and interactive visuals with HTML and SVG, and analyse images you upload. To create real images, run an image Skill in Claude Code with fal.

Which image API do I need?

Skillry image Skills use fal. Create a fal API key, set it as FAL_KEY, and the Skill handles the model calls; you pay fal directly for what you generate. You do not need a separate account with each model provider, and the same key works for every Skillry image Skill you install.

Does it work in Codex or ChatGPT?

Yes. Codex and ChatGPT can also generate images with their built-in GPT Image 2, but Skillry image Skills use fal by default in every agent, so the same FAL_KEY setup works everywhere.

What can I make with Claude image generation?

Anything an image Skill is built for: OG and social share images, event and launch posters, infographics that explain a process, diagrams, editorial illustrations and article illustration sets. Each Skill fixes the size, layout and style, so a series of images looks like it came from one designer.

Can I keep my brand colours and fonts?

Yes. Style rules live in the Skill’s SKILL.md, so you can edit the palette, typography and layout once and every image follows them. Claude applies those rules when it writes each prompt, which keeps a whole series consistent without repeating your brand guidelines in every request.

Is it free?

Some image Skills are free and others cost $4.99 once. Image generation itself is billed by fal to your own fal account, based on what you generate. Claude Code or your other agent is billed separately under your existing plan.

Make your first image with Claude and fal

Pick a style, add your fal key, describe what you need.