VigoZhao/AI-Visual-Prompt-Cookbook

What it solves

It addresses the inconsistency and lack of reusability in AI image prompting. Instead of using one-off text blobs that are difficult to iterate on or compare, this project provides a structured system for maintaining consistent visual styles across different image generations.

How it works

Each visual style is distilled into a style.json file. This file contains a prompt_template with placeholders for variables (like subject, location, and text), along with fidelity anchors and negative prompts to ensure the style remains stable. Users copy the JSON into a multimodal LLM (such as ChatGPT or Gemini), provide specific values for the variables, and the LLM generates the final detailed prompt for the image model.

Who it’s for

It is designed for designers, creators, and AI artists who use LLM-based image generation workflows and want a professional, repeatable art direction for their visuals.

Highlights

• Structured Style Library: Includes 124 curated visual styles ranging from editorial posters and zines to travel collages and minimal compositions.
• Plug-and-Play JSON: Uses a machine-readable format that can be dropped directly into LLMs to maintain style fidelity.
• Variable-Based Customization: Allows users to swap subjects and text while keeping the underlying visual system intact.
• Multimodal Optimization: Specifically optimized for models that can reason through structured JSON to produce high-quality images.