generate-image

Trigger phrases

Phrases that activate this skill when typed to Claude Code:

• generate an image
• create an illustration
• AI image generation
• make a photo of
• create visual assets

What it does

generate-image is a Claude Code skill from K-Dense AI’s scientific-agent-skills repo. It turns Claude into an image generation interface using FLUX and Nano Banana 2 AI models via OpenRouter — covering photorealistic image generation, illustrations, artwork, concept art, and general visual assets. The skill handles prompt construction, aspect ratio selection, and output file saving.

A session produces one or more generated image files at the specified dimensions and style, saved to the project directory. It is the general-purpose image generation counterpart to scientific-schematics (which is specialized for technical scientific diagrams).

When to use it

Reach for it when:

• You need a photorealistic image, concept illustration, or visual asset for a presentation, report, or communication material
• You want to generate multiple image variations for a visual concept before selecting the best one
• You need blog or documentation imagery that doesn’t involve precise technical diagrams

When not to reach for it:

• Technical scientific diagrams (flowcharts, biological pathways, neural network architectures) — use scientific-schematics
• Infographics with structured data and text — use infographics

Install

Copy the SKILL.md from K-Dense AI’s generate-image folder into .claude/skills/generate-image/ in your project. Requires an OpenRouter API key set as OPENROUTER_API_KEY.

Trigger phrases: “generate an image”, “create an illustration”, “AI image generation”, “make a photo of”.

What a session looks like

A typical session has three phases:

1. Prompt and specification. Describe the image: subject, style (photorealistic, illustration, watercolor, digital art), aspect ratio, and any specific elements to include or exclude. Claude refines the prompt for the target model.
2. Generation. The image is generated via the OpenRouter API using FLUX or Nano Banana 2. Claude selects the model based on the style request — FLUX for photorealistic outputs, Nano Banana 2 for illustrated or stylized outputs.
3. Output. The generated image is saved to the specified path and the file path is returned. Multiple variations can be generated with slightly different prompts for selection.

Receipts

Where it works well:

• Concept illustrations for scientific communication — photorealistic representations of biological processes or chemical environments where artistic interpretation is acceptable
• Blog and documentation header images where the goal is visual engagement rather than scientific accuracy

Where it backfires:

• Precise molecular structures, specific protein conformations, or accurately labeled diagrams — AI image generation produces plausible-looking but scientifically inaccurate structures for these
• Text within generated images is frequently garbled; if your image needs readable text labels, use infographics or scientific-schematics instead

Pattern that works: separate the conceptual/artistic visual elements from the text label elements — generate the artwork, then add labels in a post-processing step using matplotlib or an image editor to ensure text accuracy.

Source and attribution

Originally authored by K-Dense Inc.. The canonical SKILL.md lives in the generate-image folder of their public scientific-agent-skills repository.

License: MIT. Install, adapt, and redistribute with attribution preserved.

This page documents the skill from a practitioner’s perspective. For the formal spec and any updates, defer to the source repo.