gpt-image-2

GPT Image 2 is a focused AI skill for image generation and editing powered by the GPT Image 2 model. It operates across three distinct runtime modes to adapt to different environments: Mode A (Garden local mode) provides full end-to-end image generation by calling the OpenAI-compatible API directly and saving outputs to disk; Mode B (Host-Native delegation) acts as a prompt engineering guide when the host agent already has its own image generation tools; and Mode C (Advisor mode) degrades gracefully to a high-quality prompt consultant when no image tooling is available. This adaptive design ensures the skill remains useful regardless of the deployment context.

The skill covers 18 major categories and 80+ structured prompt templates, spanning posters, UI mockups, product visuals, infographics, academic diagrams, technical architecture charts, comics, avatars, storyboards, film storyboards, IP merchandise, and editorial workflows. It exposes two core API operations — text-to-image generation (`POST /images/generations`) and image editing (`POST /images/edits`) — and includes a lightweight mode-detection script (`check-mode.js`) that must be run first to determine the correct behavior. Prompts and images are saved under a consistent naming convention with task slugs and timestamps to prevent collisions and support version management.

This skill is designed for developers, designers, and AI agent builders who need a reliable, environment-aware image generation layer. It is particularly valuable in agentic pipelines where the execution environment may vary — from a fully configured local Garden setup with an API key, to a hosted agent platform with native image tools, to a bare environment where only prompt crafting is possible. The configurable `OPENAI_BASE_URL` also makes it compatible with third-party OpenAI-compatible gateways beyond the official API.