aldegad/image-gen

This repository provides a "skill", a small automation module, for Codex (an AI coding assistant) that generates images using Codex's built-in image generation tool. Instead of calling an external image API directly, it works through Codex's own tool by spawning a fresh, isolated Codex session each time an image is requested. The key design decision is execution isolation: image generation requests are handled in separate, short-lived sessions rather than in the same session where the caller is working. This keeps image prompts from interfering with or polluting any long-running context the main agent is managing. Authentication runs through ChatGPT's login credentials managed via Codex, not through a separate API key. After generating an image, the skill verifies the output by checking for PNG files in a specific folder under the Codex session directory, rather than trusting the file paths the model reports back, an added reliability check since models can sometimes report incorrect paths. For producing images with transparent backgrounds, the workflow uses a chroma key technique: images are first generated against a solid colour background (either bright magenta or bright green), then a helper Python script removes that background to produce a transparent PNG. Batch and continuation scripts for more complex workflows are included as experimental additions under the scripts folder. Installing the skill requires an existing Codex installation and uses Codex's own skill installer tooling. Transparent PNG post-processing requires the Python Pillow image library, and ImageMagick is noted as optionally useful for inspecting results.