seo-image-gen

Trigger phrases

Phrases that activate this skill when typed to Claude Code:

• generate OG image
• hero image
• social preview
• seo image
• generate visual

What it does

seo-image-gen is an extension skill in AgriciDaniel’s claude-seo repo. It generates production-ready images for SEO use cases using Gemini’s image generation via the banana Creative Director pipeline, with pre-configured parameters for each SEO asset type: OG/social preview images (16:9, 1K resolution, Product/UI-Web domain), blog hero images (16:9, 2K, Cinema/Editorial domain for dramatic atmospheric results), schema images (4:3, 1K, Product), product photography (4:3, 2K, white background studio lighting), and infographics (2:3, 4K, data-heavy vertical layout).

After every generation the skill runs a post-generation SEO checklist: suggests keyword-rich alt text for the image, recommends SEO-friendly file naming (keyword-description-widthxheight.webp), provides the WebP conversion command, targets file size thresholds (under 200KB for heroes, under 100KB for thumbnails), and generates the appropriate schema markup — ImageObject JSON-LD or the og:image meta tag block, depending on use case.

When to use it

Reach for it when:

• A blog post needs a hero image or OG thumbnail and you want it generated to the correct dimensions without leaving Claude Code
• You need a product photo or schema ImageObject image that matches a specific description
• You want to generate 3 variations of an OG image to A/B test social share performance

When not to reach for it:

• The banana extension is not installed (./extensions/banana/install.sh); the skill will report this and provide install instructions
• You need photography of a real physical product — AI image generation handles conceptual visuals but cannot produce accurate photos of specific physical objects

Install

This is an extension skill requiring the nanobanana MCP server. Install via:

./extensions/banana/install.sh

Then copy the seo-image-gen SKILL.md into .claude/skills/seo-image-gen/.

Trigger phrases: “generate image”, “OG image”, “social preview”, “hero image”, “blog image”, “product photo”, “infographic”, “seo image”, “create visual”, “image-gen”, “favicon”, “schema image”, “banner”, “thumbnail”.

What a session looks like

A typical session has three phases:

1. Use case identification and parameter mapping. The skill identifies the SEO image type from the command (og, hero, product, infographic, custom) or infers it from context, then applies the pre-configured aspect ratio, resolution, and domain mode for that use case. Brand presets are checked if configured.
2. Reasoning brief construction and generation. A 6-component Creative Director reasoning brief is built (Subject 30%, Style 25%, Context 15%, plus Mood, Technical, Negative weights). The skill is deliberately specific and visceral in the brief — describing what the camera sees, not abstract concepts. Generation runs via gemini_generate_image MCP tool. Approximate cost: ~$0.04/image at 1K resolution, ~$0.08 at 2K.
3. SEO checklist and schema output. The generated image path is returned along with: the crafted prompt (educational transparency), model and settings used, suggested alt text, recommended file name, WebP conversion command, and the relevant schema snippet — ImageObject JSON-LD for article images, og:image meta tags for social previews.

Receipts

Works well: OG images and blog hero images are the highest-value use cases — they require specific dimensions that differ from standard AI image prompts, and the pre-configured parameters eliminate the usual trial-and-error on aspect ratio and resolution. The post-generation SEO checklist is a genuine time-saver for people who would otherwise forget to set alt text or convert to WebP.

Backfires: Infographics with text embedded in the image are the weakest use case for current Gemini models — text rendering in AI-generated images is imprecise and may require multiple iterations. The skill acknowledges this and recommends thinking: high mode for text-heavy infographics.

Pattern that works: Generate the OG image before publishing, not after. OG images that are missing when a post first gets shared often cache as broken on social platforms — recovering from that can take days of cache clearing.

Source and attribution

Originally written by AgriciDaniel, built on the Claude Banana image generation skill. The canonical SKILL.md and supporting files live in the seo-image-gen folder of the claude-seo 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 updates, defer to the source repo.