/seo-image-gen

name: seo-image-gen
description: Generate SEO-ready images — OG/social previews, blog heroes, product shots, infographics, and favicons — at correct dimensions with SEO-aware filenames and alt text. Uses the Gemini CLI or a connected image-gen MCP/provider key when available; otherwise delivers production-ready prompts and exact specs to render elsewhere (there is no built-in renderer). Trigger when the user says "generate image", "og image", "social preview", "hero image", "product photo", "infographic", "favicon", or "create a visual".

seo-image-gen

**Family:** seo **Status:** Stable

Purpose

The generation half of image SEO (split from the audit because it carries different legal/asset/API risk). Produces images at correct dimensions with SEO-aware presets, file naming, and suggested alt text. It is **tool-aware**: it renders through whatever generator you have — preferring the Gemini CLI — and, when no generator is available, still delivers production-ready prompts + exact specs + filenames + alt text so you can generate elsewhere. That prompts-and-specs output is real value on its own; it never just fails.

Triggers

• "generate image" / "create a visual"
• "og image" / "social preview"
• "hero image" / "blog image"
• "product photo" / "infographic" / "favicon"

Inputs

• Subject/prompt and brand context (palette, style — pull from a design system)
• Preset: og | hero | product | infographic | favicon
• Output path (in the user's project workspace)

Steps

1. **Pick the preset** and resolve its spec — e.g. OG 1200×630; hero ~1600×900; favicon set (16/32/180/512). Pull palette/style from a `design-system-gen` result so generated images match the brand. 2. **Compose the generation prompt** — subject + style + palette + composition, with negative prompts to avoid unwanted text artifacts. 3. **Detect generators.** Run `python3 "${CLAUDE_PLUGIN_ROOT}/scripts/workflow/capability_probe.py"` and check for a connected image-gen MCP. Route per the cascade below. 4. **Generate or hand off.** If a generator is available, render at the preset dimensions into the user's workspace. If none is, output the prompt + exact specs and stop with a one-line note on how to enable rendering (install the Gemini CLI or connect an image-gen MCP/provider key). 5. **Name + alt.** SEO-friendly filename (kebab-case, descriptive) + suggested alt text per image. 6. **Hand to** `seo-image-audit` to verify generated assets pass the audit.

Capability routing

This skill follows the plugin's capability-tier cascade (`references/CAPABILITY-TIERS.md`). For image generation there is no built-in Claude renderer, so the numbers run 3 → 1 → 4 by design (the CLI is preferred over an MCP):

1. **Tier 3 — Gemini CLI (preferred).** If `gemini` is on PATH (per `capability_probe.py`), use it to render at the preset dimensions. 2. **Tier 1 — image-gen MCP / provider key.** Otherwise, if an image-gen MCP is connected or a provider key is set (see `extensions/nanobanana/`), render with it. 3. **Tier 4 — guided (always available).** If neither is present, deliver the production-ready prompts + exact specs + filenames + alt text, and offer the options to enable rendering (install `@google/gemini-cli`, or connect an image MCP).

Always state which tier ran and how to enable rendering if it didn't.

Outputs

• Generated image(s) at preset dimensions (Tier 3 or Tier 1)
• Otherwise: ready-to-use prompts + exact specs + SEO filenames + suggested alt text
• A one-line "to render here, install/connect …" note when handing off

Dependencies

• Optional generator (one of): the Gemini CLI (`@google/gemini-cli`, preferred), or

an image-gen MCP / provider key (`extensions/nanobanana/`). Required only to actually render; the prompts+specs path works without any.

• `design-system-gen` (optional — for brand-matched output)
• `scripts/workflow/capability_probe.py`

Notes

You are responsible for the commercial rights to generated images; review the model's terms (see `extensions/nanobanana/README.md`). Pairs with `seo-image-audit`.