gpt-image2
GPT Image 2
This skill lets Codex behave like a drawing assistant that can call a gpt-image-2 compatible image service. It first tries to inherit OpenAI/Codex-compatible environment credentials already exposed to the current process, then reads the current Codex provider base URL from ~/.codex/config.toml and the Codex OPENAI_API_KEY from ~/.codex/auth.json, and it also supports local GPT_IMAGE2_* overrides. It is for direct generation from the current machine: interpret the user's visual request, run the bundled CLI, save the returned image files, and report the local paths.
Use the existing imagegen skill when the user needs the official OpenAI Image API workflow with editing, masking, background work, or broader project asset generation. Use this skill when the user specifically wants this gpt-image-2 compatible drawing path, inherited Codex/OpenAI-style credentials, a configured URL/sk pair, or a simple "Codex, draw this for me" workflow backed by scripts/gpt-image2.mjs.
When to Use
- The user asks Codex to draw, generate, create, render, or make a new image through the configured drawing service.
- The user mentions
gpt-image-2,gpt-image2, inherited Codex/OpenAI credentials, a local drawing channel, URL/sk configuration, or the includedgpt-image2.mjsscript. - The request is text-to-image with optional reference images, image count, resolution, aspect ratio, quality, and output directory.
- The user wants saved local image files rather than prompt advice or sample API code.
- The user asks to validate request shape with a dry run or check an official-compatible image API mode.
Do not use this skill for mask-based editing, local canvas repainting, inpainting, background removal, or precise retouch workflows. Route those to a more suitable image editing skill if one is available.