GPT-image2 Image Generation

Problem

When you want to move image generation from a web UI into scripts, pipelines, or batch jobs, calling OpenAI's gpt-image-2 directly means handling authentication, task submission, polling, and file output. This skill condenses that flow into a local script: provide a prompt to generate an image, or use a reference image for editing. It is useful for terminal-based prototyping, asset variants, and automated image generation.

How It Works

• Text-to-image: use prompt to describe the scene and output PNG, JPEG, or WebP.
• Image-to-image: pass a reference image with --image, then use --fidelity high/low to decide whether to preserve detail or allow larger changes.
• Parameterized output: supports whitelisted sizes, up to 10 images per run, --bg transparent for transparent backgrounds, and --compression for quality control.
• Task control: use --no-download to submit without waiting, and --task-id to inspect existing tasks, which helps integrate generation into longer workflows.

The script submits the request, waits for completion, and saves the result to the configured output directory.

Boundaries

• Use only the listed fast and high-resolution sizes; custom resolutions may be slow or fail.
• prompt is limited to 500 characters; reference images are currently PNG, JPEG, or WebP.
• Transparent output requires PNG or WebP; JPEG does not support transparency.
• Generation usually takes about 10–30 seconds per image, with longer waits for complex scenes or batch jobs.