How to use the Image Generation (Qwen-Image) agent skill

What an agent skill is and what image-gen does

An agent skill is a folder with a SKILL.md file that gives an AI agent instructions for a specific task. When your request matches the skill's description, the agent follows those instructions.

When you ask for a picture, poster or illustration, the image-gen skill has the agent write a short Python script. The script sends your description to SiliconFlow's Qwen-Image model by default, reads the image address from the response, downloads the image inside the sandbox, and saves it to your project folder. The agent then shows the image in the chat and, if you want a downloadable file, registers it as an artifact.

• Default endpoint: POST https://api.siliconflow.cn/v1/images/generations
• Default model: Qwen/Qwen-Image
• Default parameters: negative_prompt is an empty string, num_inference_steps is 20, guidance_scale is 7.5

Install the image-gen skill

Download the ZIP for this skill from this page. Then choose one of the options below.

For a personal setup, unzip the download into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal: unzip into ~/.claude/skills/
• Project only: unzip into <your-project>/.claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up the requirements

The skill needs a SiliconFlow account and an API key. The key must be set as the environment variable SILICONFLOW_API_KEY in the agent sandbox environment. The skill reads it there and does not rely on the backend server's environment variables.

The sandbox also needs Python with the requests library. The skill depends on two other pieces: the chat-widgets skill, which displays images in the chat, and the present_artifacts tool, which creates downloadable file cards. Without SILICONFLOW_API_KEY the skill cannot produce images, and the agent is instructed to tell you to add it.

• SiliconFlow account and API key
• SILICONFLOW_API_KEY set in the agent sandbox environment
• Python with the requests library in the sandbox
• chat-widgets skill (dependency)
• present_artifacts tool (dependency)

Trigger the skill with natural requests

You do not need a special command. Ask for an image in plain language. The skill is meant to activate when you ask to generate an image, poster, illustration or text-to-image output, or when you name Qwen-Image or another compatible image interface.

If you don't give size, style or other details, the agent is told to use reasonable defaults rather than asking you many follow-up questions. You can add detail when you want more control.

• Generate an illustration of a quiet harbour at dusk in soft watercolour style.
• Make a poster for a community bake sale, warm colours, room for a title at the top.
• Use Qwen-Image to create a flat vector icon of a bicycle for my app's about page.
• Save the image into this project's assets folder as a PNG.

Check the results and where files are saved

The agent saves the final image into the current working folder in a location that fits your project structure, unless you say where to put it. The reply should show the image in the chat. The agent is instructed not to show the temporary image address from the API as the final result.

If you need to download or keep the file, ask for a downloadable file. The agent then registers the saved path with present_artifacts so you get a file card.

• Saved as generated-image.png in the project folder in the default example script
• Final reply displays the image inline through the chat-widgets Image format
• Downloadable file card appears only when you ask for a file

Safety notes: scripts and network use

This skill runs scripts and uses the network. Your description is sent to the third-party API at api.siliconflow.cn, and the generated file's content leaves your environment for that service. Do not use it for sensitive or confidential prompts.

Generated images can be unpredictable. For final brand assets or anything that needs fully controlled output, review the result by hand. The skill also tells the agent not to call the backend MinIO upload tool; image generation and download both happen inside the sandbox.

• Prompts are sent to api.siliconflow.cn
• Do not use for confidential or sensitive content
• Review outputs before using them as final brand assets
• Image generation and download happen inside the sandbox

Use it with other compatible image models

If you name another image model or compatible interface, the agent can call it following that interface's protocol. Whatever the method, the goal is to get the image bytes or base64 data, save the file in the working folder, and then display it using the chat-widgets Image format.

The default example is a template. The agent can adapt its prompt and save location, so you can use it as a starting point for your own API calls.