GPT-Image2-Skill

What GPT-Image2-Skill does

The repository combines a gallery of 163 numbered prompt entries across 31 categories with reference images and task-specific guides. Two skills do the agent-facing work: one handles image generation and editing, the other extracts a prompt from a reference image so a vision-capable agent can pass it to a generator. The image-to-prompt skill can be called with a slash command, a $get-prompt-from-image call or plain language.

A CLI accepts reference images with the -i flag, and the README advises exporting PDF pages as PNG, WebP or JPG first. Its own guidance is to keep prompts short, make the reference clear, and state required text and edit constraints explicitly. The Image API returns PNG, JPEG or WebP, so editable SVG or PowerPoint shapes need a separate authoring step. Documentation is in English and Chinese, and the surfaces list names Claude Code, Codex, OpenClaw, Hermes Agent and other skill-capable runtimes.

Key features

• 163 numbered prompt gallery entries across 31 categories
• Skill for GPT Image generation and editing
• Get-prompt-from-image skill that extracts a prompt from a reference image
• CLI with reference-image input through the -i flag
• Sample images generated with GPT Image 2.5 at 2048x2048
• English and Chinese documentation

When to use it

• Browsing worked prompts for technical illustrations, isometric scenes and similar styles
• Letting an agent generate or edit an image from a reference
• Turning an existing image into a reusable prompt

Who it is for: Developers and designers who direct image generation from an AI agent and want tested prompts to start from.

How it fits with Hermes Agent

Supports Hermes Agent among other skill-capable runtimes, including Claude Code, Codex and OpenClaw. It is a general image-generation skill pack, not built only for Hermes.

FAQ