gemini-creative-plugin

Gemini Creative Plugin

A Claude Code plugin wrapping a Model Context Protocol (MCP) server for AI image generation using Google's Gemini 3 image models. Designed for LLM agents — built around sessions, organized file output, and 9 auto-activating skills covering game art, UI mockups, character sprites, and prompt engineering.

Version: v0.5.1 · Default model: Gemini 3.1 Flash (Nano Banana 2) · Also supports: Gemini 3 Pro · License: MIT

---

Install

Two slash commands inside Claude Code:

/plugin marketplace add https://github.com/2901were/gemini-creative-plugin
/plugin install gemini-creative-plugin@gemini-creative-marketplace

Then set your Gemini API key (get one at aistudio.google.com/apikey):

export GEMINI_API_KEY="your-key-here"

That's it. The plugin registers the MCP server and the 9 auto-activating skills. Restart Claude Code after setting the env var (MCPs inherit env at launch).

---

What it does

10 MCP tools

Image generation:

• generate_image — text-to-image with optional reference images (max 14, identity/style anchors), aspect ratio, resolution, and thinking-level controls
• edit_image — modify an existing image with text + up to 14 reference images
• continue_editing — iterate on the last image without re-specifying its path

Sessions:

• start_creative_session / send_creative_message / end_creative_session — organize multi-image workflows with shared defaults (model, aspect ratio, output directory)
• get_session_info — list active sessions or inspect one

Setup + utilities:

• configure_gemini_token / get_configuration_status — API-key setup
• get_last_image_info — file path, size, and timestamp of the last generated image
• get_prompt_template — pre-built prompt structures for common asset types

9 auto-activating skills

Claude Code triggers these automatically when the conversation matches their description:

Skill	Triggers on
/gemini-quick	Single image, fast one-off
/gemini-characters	Game characters — style bible, poses, cross-session consistency
/gemini-sprite-series	Consistent character sprites across poses or animation frames
/gemini-game-assets	Pixel art, tilesets, palettes, isometric, tier evolutions
/gemini-ui-mockups	Multi-screen UI mockups with style selection
/gemini-session	Session setup, model selection, troubleshooting
/gemini-workflows	Workflow decision tree across the four main patterns
/gemini-prompts	Domain-specific prompt structures (sprites, UI, icons, web)
/gemini-configure	API-key setup and authentication troubleshooting

---

Best practices

For visual consistency across multiple images, always pass the previous image via the images: [path] parameter in send_creative_message. Without it, Gemini reconstructs from text alone and details drift between calls — proportions shift, colors change, fine features vary.

Prompt structure:

[Style] [Subject] in [pose/state], [composition], [distinctive details], [background], [technical specs]

Pixel-art bible prefix:

2D pixel art, 32x32 sprite, PICO-8 16-color palette, 1px black outline, 2-tone cel shading, transparent background —

---

Troubleshooting

"Gemini API token not configured" — Run get_configuration_status to verify setup. Set GEMINI_API_KEY env var and restart Claude Code (MCPs inherit env at launch).

Images generating inconsistently across multiple calls — Pass the previous image via the images: [path] parameter in send_creative_message. The guide-pinning approach + image references are what locks consistency.

"Maximum sessions (5) reached" — Run get_session_info and end an older session.

Array parameters not working — Ensure your MCP client sends arrays as JSON arrays, not strings. The server handles both formats defensively, but the JSON form is canonical.

---

Links

• Get a Gemini API key — free tier available
• Gemini API documentation
• Image generation guide
• Model Context Protocol

---

Built by Vladimir Bulanenko. MIT licensed.