Prompt Grafting: extracting style-only modular prompts from images, works across LLMs

TheChuckTone · x · 2026-09-06

Designer DtheDesigner shares a discovery called "Prompt Grafting": strip the specific content out of an image or prompt, leaving a modular block of style and layout guidance you can reuse across generations. He's packaging it as a Markdown file (Prompt Grafter) on GitHub and a ChatGPT plugin called Prompt Graft. The technique appears platform-agnostic — any LLM that can analyze images should work — and he plans to test it in Claude, Gemini, and Grok. He uses it alongside Adobe Firefly Boards and source images, calling the results remarkable.

Original post →

More from coding & agent

coding & agent channel →