gpt-image

gpt-image is a thin integration skill for generating and editing images with OpenAI's GPT-Image-2 model through the inference.sh CLI (the belt tool). It solves the task of driving a single model that handles text-to-image generation, reference-image editing, and mask-based inpainting from the command line, with batch generation and flexible output formats.

The skill documents the belt workflow: belt login to authenticate, then belt app run openai/gpt-image-2 with a JSON input payload. It provides worked examples for text-to-image with quality settings, generating multiple images at once (up to 10), editing with one or more reference images, multi-image composition, mask-based inpainting, custom resolutions in 32px increments from 256 to 4096, and fast low-quality drafts. A parameter table documents prompt, images, mask, n, quality, width, height, output_format, and output_compression, and a pricing table lists approximate per-image costs by quality tier. It also lists related inference.sh skills for other model families and points to platform documentation.

Target users are developers and creators who want scriptable access to GPT-Image-2 for product mockups, marketing visuals, concept art, photo manipulation, and inpainting. It depends on the external belt CLI (installed separately via npx skills add belt-sh/cli) and a paid inference.sh account. The skill is restricted to running the belt CLI and performs only image-generation operations against a hosted service — no destructive local operations, credential exfiltration, or unsafe execution.