What problem does it solve? Editing images with AI models often produces drift: faces change, brand marks get rewritten, and in-image text comes out wrong. This Skill packages the documented prompting patterns for OpenAI GPT Image 2's /edit endpoint on RunComfy so edits preserve identity, layout, and multilingual text instead of relying on naive prompting. ## Core Features & Use Cases - Preservation-first editing: Lead prompts with preservation goals (face, pose, brand, framing) so targeted changes don't alter the rest of the image. - Multilingual in-image text rewriting: Quote exact characters and name the script (Japanese kana, Cyrillic, Arabic, CJK) to swap headlines and labels across language variants. - Multi-reference composition: Pass up to 10 image URLs and reference them by number to combine subject, scene, lighting, and palette from different sources. - Use Case: Localize an ad poster into Japanese by keeping the photograph, layout, and brand mark unchanged while replacing only the headline with quoted kana text. ## Quick Start Ask the AI to edit an image with GPT Image 2 on RunComfy, for example: replace the background of this portrait URL with a soft grey studio sweep while keeping the person's face and pose unchanged.
