gpt-images

# GPT Images Input: a brief in plain language, one sentence is enough. Output: an English prompt ready to paste, a settings block for the chosen surface, and two or three levers that actually change the result. `v1.0.0`. Self-contained: this skill does not depend on any other skill being installed. **Generate nothing on your own.** Image generation costs the operator money. Produce the prompt, then ask. Hard numbers (model IDs, sizes, quality levels, prices, limits) live in `references/official-spec.md`. Never quote them from memory, read the file. ## Step 0: pick the surface Always first. Without a surface there is no settings block. | Surface | When | What you deliver | |---|---|---| | **ChatGPT app** (default when nothing else is said) | manual work, iteration, sketch input, built-in templates | the prompt in plain language, no API parameters | | **OpenAI API** (`v1/images/generations`, `/edits`, or the Responses API tool) | scripts, pipelines, transparent backgrounds, exact pixel sizes | prompt plus `model`, `size`, `quality`, `background`, `output_format` | | **Third-party hosted surface** | a platform that resells GPT Image models | prompt plus whatever that platform's live catalog exposes. Check the catalog, do not assume it matches the API | ## Step 1: intake, five things 1. **Purpose and channel.** This decides the size. A social post, an article cover and a print poster are three different jobs. 2. **Subject.** What is actually in the frame. If a specific p...