Skip to content
Get packs

Guide · Agent skill

How to use the Image Generation agent skill (image-generation)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The image-generation skill lets an AI agent turn a text request into a picture, such as a character concept, scene, or product visual. It writes a structured JSON description of what you asked for, then runs a bundled Python script that sends that description to an image service and saves the result as an image file. This guide explains what the skill does, how to install it, how to trigger it, what it needs, and what to watch for.

Image Generationbytedance/deer-flow · MIT

Quick start

  1. 1Check requirements. Confirm Python is installed with the requests and Pillow libraries, and that you have an API key for MiniMax or an OpenAI-compatible service.
  2. 2Download and install the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
  3. 3Set your API key. Provide your provider credentials as the skill expects, using MINIMAX_API_KEY for MiniMax or your OpenAI-compatible key.
  4. 4Describe the image. Ask the agent in plain language for a character, scene, or product visual, and include the style, mood, and any reference images.
  5. 5Review the output. The agent saves the image, typically under /mnt/user-data/outputs/, and describes the result.
  6. 6Iterate. Ask for changes if the result is not right. The skill offers to iterate when adjustments are needed.

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md instruction file and, optionally, scripts and templates. When a request matches the skill's description, the agent follows the instructions in SKILL.md instead of improvising.

This skill is triggered when you ask to generate, create, imagine, or visualize images, including characters, scenes, products, or other visual content. It supports structured prompts and reference images for guided generation.

  • Creates a structured JSON prompt file describing subject, style, composition, lighting, and colour palette
  • Runs scripts/generate.py to send the prompt to an image-generation service
  • Accepts reference images so the output can follow their look or composition
  • Saves the result as an image file, by default under /mnt/user-data/outputs/ in the examples

How do you install the image-generation skill?

Download the ZIP of the skill from this page. Then either unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder so that it applies only to that project.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into the project's .claude/skills/ folder
  • Claude apps: Settings → Capabilities → Skills, then upload the skill

What do you need before the skill can run?

The skill cannot generate anything without credentials. You need Python with the requests and Pillow libraries, an API key, and network access to the image provider.

The script supports MiniMax (set MINIMAX_API_KEY) or an OpenAI-compatible image service. The OpenAI endpoint is the default.

  • Python with the requests and Pillow libraries installed
  • An API key for MiniMax (MINIMAX_API_KEY) or an OpenAI-compatible image service
  • Network access to the image provider

How do you trigger the skill with a natural request?

You do not need special syntax. Describe the image you want in plain language and say that you want it generated, created, or visualized. The agent matches the request to the skill's description and follows its workflow.

Include the details that the skill asks about: the subject, the style or mood, the aspect ratio or composition, and any reference images you want it to follow.

  • Ask for a character concept: "Create a Tokyo street style woman character in the 1990s"
  • Ask for a scene: "Visualize a rainy harbour town at dusk, painted in muted blues"
  • Ask for a product visual: "Generate a product image of a ceramic mug on a wooden table, soft side light"
  • Add references: "Generate a scene using the character in my uploaded picture"

What happens when the skill runs?

The agent first identifies the subject, style, technical specs, and any reference images. It then writes a JSON prompt file in /mnt/user-data/workspace/, named after the content, and calls the script with the prompt file, an output path, and optionally reference images and an aspect ratio.

The script parameters are --prompt-file (required, absolute path to the JSON file), --reference-images (optional, space-separated absolute paths), --output-file (required, absolute output path), and --aspect-ratio (optional, default 16:9).

  • --prompt-file: absolute path to the JSON prompt file (required)
  • --reference-images: absolute paths to reference images (optional)
  • --output-file: absolute path to the output image file (required)
  • --aspect-ratio: aspect ratio of the output (optional, default 16:9)
python /mnt/skills/public/image-generation/scripts/generate.py --prompt-file /mnt/user-data/workspace/cyberpunk-hacker.json --output-file /mnt/user-data/outputs/cyberpunk-hacker-01.jpg --aspect-ratio 2:3

What does a structured prompt look like?

The prompt is a JSON file. The skill's character example includes fields such as characters, negative_prompt, style, composition, lighting, and color_palette. Other scenarios, such as scenes and product visuals, use different fields, as the skill describes Scene Generation and Product Visualization schemas.

When you use reference images, the prompt refers to them as [Image 1] and [Image 2] in the same order as the files you pass to --reference-images.

  • characters: gender, age, ethnicity, body type, facial features, clothing, accessories, era
  • negative_prompt: what to avoid
  • style, composition, lighting, color_palette: visual direction
  • technical: optional aspect_ratio, quality, and detail_level
python /mnt/skills/public/image-generation/scripts/generate.py --prompt-file /mnt/user-data/workspace/star-wars-scene.json --reference-images /mnt/user-data/uploads/character-ref.jpg /mnt/user-data/uploads/vehicle-ref.jpg --output-file /mnt/user-data/outputs/star-wars-scene-01.jpg --aspect-ratio 16:9

What about the Doraemon comic template?

The skill includes a template for an 8-panel Doraemon comic. The agent reads it only when the request matches, for example when you ask for a Doraemon comic. The template describes a 1080 by 1920 canvas, an 8-panel layout in two columns, and the characters Doraemon, Nobita, Shizuka, Giant, and Suneo.

Use it when you want a story told in panels. For other subjects, the general workflow is enough.

  • Workflow: extract the story context, map it to 8 narrative beats, then fill in the JSON template
  • Panel layout: two columns, four rows, with left and right column positions given in the template

What are the safety and privacy notes?

This skill runs a Python script and uses the network. The script sends your prompt, and any reference images you supply, to the image provider you configured. Review what you upload before you run it, and do not send content you are not allowed to share.

The skill is built to closely match uploaded reference images. Do not use it to generate images of real people from reference photos without their consent.

  • Runs scripts: yes. Uses network: yes.
  • Reference images leave your machine and go to the configured provider
  • Keep API keys out of shared prompts and files

What people use it for

  • Drafting character concepts from a text description
  • Exploring mood boards or scene art before commissioning final work
  • Mocking up illustrative images for posts or slides
  • Producing a multi-panel Doraemon-style comic using the included template

Frequently asked questions

How do I install the image-generation skill?

Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.

What do I need to run the image generation skill?

You need Python with the requests and Pillow libraries, an API key for MiniMax (MINIMAX_API_KEY) or an OpenAI-compatible image service, and network access to that provider. The script cannot run without credentials.

Is the image generation skill free?

The source does not state pricing or licence fees for the image provider. The skill's code is listed under the MIT licence. Check your provider's own pricing and terms before you use it.

Which image provider does it use?

It supports MiniMax, set with MINIMAX_API_KEY, or an OpenAI-compatible image service. The OpenAI endpoint is the default.

Can I use reference images to guide the output?

Yes. Pass one or more reference images with --reference-images, and refer to them in the prompt as [Image 1] and [Image 2]. The output will try to follow their look or composition, so do not use it to imitate real people without consent.

Why does my image not match the description?

The output comes from an external model and may vary or be inaccurate. Make the JSON prompt more specific, add a negative_prompt for things to avoid, and check that your reference images and file paths are correct.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you want character concepts, scene art, or product mood visuals drafted quickly from text, and you have an API key for MiniMax or an OpenAI-compatible service. Pick a different tool if you need exact, legally sensitive artwork, or if you cannot send prompts and reference images to an external provider.

Download the skill