VigoZhao/AI-Visual-Prompt-Cookbook

What it is

AI Visual Prompt Cookbook is a community‑maintained library of visual‑style prompts for text‑to‑image generation. Each style lives in its own folder as a single style.json file that contains:

• a prompt template with placeholders like {SUBJECT} or {ASPECT_RATIO}
• a list of environment variables that define which placeholders are required
• optional example cases showing concrete values
• “fidelity anchors” and a negative prompt that tell the model what to keep or avoid.

The idea is to make sophisticated, repeatable visual directions easy to drop into any LLM‑based image model (ChatGPT‑4‑Vision, Gemini‑3‑Pro Image, etc.). Users copy the JSON, fill in their own values, paste the whole thing into the chat, and the model returns a fully‑formed image prompt.

The repo ships 169 ready‑made styles (e.g., “Everyday Riso”, “Chromatic Gesture Characters”, “Velocity Type Sport”) with preview images and short copy‑ready prompts. Styles are grouped by visual theme (photo + doodle, zine + collage, type posters, travel, editorial, product). An online gallery and a discussion space let users showcase results and request new styles.

---

Who might use it

Audience	Why it helps
Creative professionals (graphic designers, art directors)	Quickly generate consistent mock‑ups or mood‑boards without hand‑crafting long text prompts each time.
Prompt engineers / AI artists	A structured JSON format makes it easy to version‑control, share, and iterate on visual systems across projects.
Developers building multimodal pipelines	The style.json can be program‑matically loaded, variables filled, and the final prompt sent to an image‑generation API.
Educators / students	Provides concrete examples of how to decompose a visual style into reusable components.

---

How to get started (quick walk‑through)

1. Pick a style – Browse the featured gallery or the full catalog and open its folder, e.g. styles/mono-noir-type-portrait-poster-style/.
2. Copy the JSON – Grab style.json (or the short copy‑prompt from docs/copy-prompts/).
3. Supply values – Either edit the examples[*].values section or provide your own values for the variables listed under environment_variables.
4. Paste into a multimodal LLM – Send the whole JSON to ChatGPT‑4‑Vision, Gemini‑3‑Pro Image, or any other model that accepts long structured prompts.
5. Receive the generated image – The model will interpret the filled‑in template and output an image matching the described visual system.

The README includes a complete example that shows the template, variable values, the final assembled prompt, and a preview of the generated image.

---

What’s inside (repo layout)

AI-Visual-Prompt-Cookbook/
│
├─ styles/                     # One folder per visual style
│   ├─ <style‑slug>/
│   │   ├─ style.json          # Machine‑readable prompt definition
│   │   ├─ preview-16x9.jpg    # Landscape example
│   │   └─ preview-9x16.jpg    # Portrait example
│   └─ …
│
├─ docs/
│   ├─ CATALOG.md             # Full list of 169 styles with descriptions
│   └─ copy-prompts/          # Human‑readable, ready‑to‑paste prompts
│
├─ assets/                     # Hero image, thumbnails, etc.
│
├─ scripts/validate-style-json.py   # Simple validator for the JSON schema
│
└─ README.md (this file)

Each style.json follows version 2.2 of the schema:

{
  "schema_version": "2.2",
  "prompt_template": "Create a {ASPECT_RATIO} …",
  "environment_variables": ["ASPECT_RATIO", "SUBJECT", "MAIN_TEXT", …],
  "examples": [{"case_name": "demo", "values": {"SUBJECT": "…"}}],
  "style_fidelity_anchors": "…",
  "negative_prompt": "no watermarks, no logos, …"
}

---

How it fits into the AI ecosystem

• Prompt engineering – Turns the often‑unstructured art of writing image prompts into a repeatable, version‑controlled data format.
• Multimodal LLMs – Works with any model that can ingest a long JSON string and generate images (OpenAI’s GPT‑4‑Vision, Google Gemini‑3‑Pro Image, etc.).
• Creative pipelines – Can be scripted: a program reads a style, fills variables from a database or UI, calls the image‑generation API, and stores the result, enabling batch production of stylized assets.
• Open‑source community – New styles are added via pull‑requests, and the repository includes validation scripts to keep the JSON schema consistent.

---

When to consider alternatives

• If you need pixel‑level control (e.g., custom diffusion fine‑tuning) rather than prompt‑level style guidance.
• When working with non‑LLM image generators that don’t accept textual prompts (e.g., pure diffusion pipelines without a language model).
• If you prefer a graphical UI for prompt building; this repo is text‑centric.

---

TL;DR

The AI Visual Prompt Cookbook is a curated collection of 169 JSON‑based visual style templates for text‑to‑image models. Copy a style.json, fill in a few variables, paste it into a multimodal LLM, and get a consistent, high‑quality image that follows a predefined visual system. Ideal for designers, prompt engineers, and anyone who wants repeatable, shareable image‑generation prompts.