How to Generate Images in Claude Code (2026)

Claude Code is fantastic at writing code, fixing bugs, and explaining things. But ask it to draw you a picture and it'll politely tell you it can't — it's a text-based tool with no built-in image generation.

Here's the good news: you can teach it. In about ten minutes, you can set Claude Code up with a third-party image API, and from then on you just say "generate an image of a sunset over mountains" and it handles everything. Your API key stays tucked away in a file Claude reads but never repeats back to you.

We'll use Google's Gemini API as our example in this guide, but nothing here is Gemini-specific — OpenAI, Stability AI, Replicate, or any other image API with an HTTP endpoint works exactly the same way. Swap the key and the endpoint, keep everything else.

This guide assumes you've never done anything like this before. Let's go step by step.

Google's Gemini API can generate images. To use it (or any other provider — the steps are the same idea everywhere: sign up, create a key, copy it):

A note on pricing: Google AI Studio has historically offered a free tier, but what's included can change — check their current pricing page to confirm image generation is covered before you start generating dozens of images. The same goes for any provider: know what each image costs you before you scale up.

Treat this key like a password. Anyone with your key can generate images on your dime (or burn through your free quota). You'll never paste it into a chat window — not even to Claude.

A.env file is just a plain text file that holds secrets like API keys. Programs read it; humans rarely open it. Keeping your key here instead of in your code or chat history is standard practice.

GEMINI_API_KEY=your-actual-key-goes-here Enter fullscreen mode Exit fullscreen mode Replace your-actual-key-goes-here with the key you copied from AI Studio. No quotes, no spaces around the equals sign.

One more critical step: if your project uses git, make sure.env never gets committed. Create (or open) a file called.gitignore in the same folder and add this line:

.env Enter fullscreen mode Exit fullscreen mode That's it. Your key now lives in one place, out of your code and out of your chat history.

Here's where the magic happens. Open Claude Code in your project folder and send it a message like this:

I put my Gemini API key in the.env file in this project as GEMINI_API_KEY. From now on, whenever I ask you to generate an image, use that key with the Gemini image generation API to create it. Read the key from the.env file yourself — never ask me for it, and never print it, display it, or paste it into the chat.

That's the whole instruction. You're establishing three rules:

Claude Code will confirm it understands. From this point on, the setup is permanent for this project — you only do this once.

If you want this to survive across sessions without re-explaining, add the same instruction to a CLAUDE.md file in your project root. Claude Code reads this file automatically every time it starts. Just create CLAUDE.md and write:

## Image Generation - The Gemini API key is stored in `.env` as `GEMINI_API_KEY`. - When I ask for an image, generate it using the Gemini image API (model `gemini-3.1-flash-image` ) with that key. - Never display, repeat, or log the API key. Enter fullscreen mode Exit fullscreen mode Now every future Claude Code session in this folder knows the deal from the start.

Generate an image of a cozy cabin in the woods at sunset, warm light in the windows.

Claude Code will get to work. Here's roughly what happens behind the scenes:

A few seconds later, you'll have an actual image file sitting in your folder. Open it and take a look.

Now generate a logo concept for a coffee shop called "Moonbeam Roasters" — minimalist, crescent moon, warm browns.

Each image lands in your project folder, ready to use.

You don't need to understand this to use it, but it's nice to know what's happening — and it demystifies why this approach is so flexible.

When you ask for an image, Claude Code typically writes a short Python script that looks something like this:

import os import base64 import requests from dotenv import load_dotenv load_dotenv () api_key = os. getenv ( " GEMINI_API_KEY " ) response = requests. post ( " https://generativelanguage.googleapis.com/v1beta/models/ " " gemini-3.1-flash-image:generateContent?key= " + api_key, headers = { " Content-Type ": " application/json " }, json = { " contents ": [{ " parts ": [{ " text ": " a cozy cabin in the woods at sunset " }]}], " generationConfig ": { " responseModalities ": [ " IMAGE " ]}, }, ) data = response. json () for part in data [ " candidates " ][ 0 ][ " content " ][ " parts " ]: if " inlineData " in part: with open ( " cabin-sunset.png ", " wb " ) as f: f. write ( base64. b64decode ( part [ " inlineData " ][ " data " ])) Enter fullscreen mode Exit fullscreen mode The script loads your key from.env, sends your prompt to Gemini's image model, gets back image data, and saves it as a PNG. Claude handles all of this for you — you'll usually never need to write or even look at this code. But now you know there's no magic involved, just an API call.

The beauty of this setup: it works with any image API, not just Gemini. If you'd rather use OpenAI's image model, Stability AI, or anything else with an HTTP API, the same pattern applies — key in.env, instruction to Claude, done.

This deserves its own section because it's the part beginners worry about most — rightly so.

Here's what protects your key in this setup:

One thing to watch: if you ever ask Claude to "show me the script you used," the script it shows you will reference os.getenv("GEMINI_API_KEY") — the variable name, not the actual key. That's exactly what you want.

Once the basics work, these small habits make a big difference:

"Claude says it can't find the key." Make sure the.env file is in the same folder where you're running Claude Code, and that the variable is named exactly GEMINI_API_KEY (all caps, underscores). Check for typos and stray spaces.

"The API returns an error." The most common cause is an invalid or expired key. Go back to AI Studio, generate a fresh key, and update your.env. Also check your usage against your plan's quota — if you've hit a limit, that's your culprit.

"The image looks wrong." That's a prompting issue, not a setup issue. Try describing what you want in more detail — subject, style, lighting, mood. Image models are literal; they draw what you describe.

"Claude printed my key!" This shouldn't happen if you gave the instruction from Step 3. If it ever does, immediately go to AI Studio and revoke/regenerate your key, update.env, and remind Claude of the rule.

That's the whole setup: key from AI Studio, one line in.env, one instruction to Claude Code. Ten minutes of setup, and from then on generating images is as easy as describing what you want.

What I love about this approach is that nothing is locked in. Swap Gemini for another provider later? Just change the key and tell Claude which API to call. The pattern — secrets in.env, instructions in plain English — works for any API you want Claude Code to use, not just image generation.

Now go make something. Start with that cabin at sunset.