🧠 AI Prompt Builder

The Luminance Master - Craft perfect prompts for AI image generators

A zero-dependency, single-file web app that builds optimized prompts for Midjourney, FLUX.1, Stable Diffusion, and DALL-E 3.

🚀 Try it live →

Report Bug · Request Feature · Discussions

🌐 Live Demo

👉 https://tekideki.github.io/AI-Prompt-Builder/

No installation needed. Works in any modern browser (Chrome, Firefox, Safari, Edge).

📖 About

AI Prompt Builder is a browser-based tool that helps you craft high-quality, structured prompts for AI image generators. Instead of guessing what words work, you select from curated dropdowns covering lighting, composition, camera optics, film stock, atmosphere, and more - and the tool assembles a coherent, engine-specific prompt.

The app runs 100% client-side. No build step, no backend, no tracking. All data stays in your browser via localStorage / sessionStorage.

✨ Features

🎯 Multi-Engine Support - Midjourney, FLUX.1, Stable Diffusion, DALL-E 3
🎨 14 Built-in Presets - Cinematic Portrait, Epic Landscape, Neon Cyberpunk, Vintage Film, and more
💾 Custom Presets - Save your favorite configurations with one click
📦 18 SD Models Included - Each with trigger words, VAE info, and recommended CFG / steps / sampler
🎲 Smart Randomize - Context-aware randomization that respects your subject
🚫 Conflict Detection - Warns when settings contradict each other
📊 Live Prompt Score - Weighted 0-100 rating with visual feedback
📜 Prompt History - Last 20 prompts, click to restore & copy
🗂️ Batch Processing - Upload a .txt of subjects → generate all prompts at once
🎯 Multi-Engine Batch - Generate the same prompt for all 4 engines simultaneously
📤 Export SD Batch - Download all queued prompts as a single .txt
📥 JSON Import/Export - Settings, history, custom presets, custom models and theme, with strict validation (prototype-pollution safe)
🎭 AI Analysis Panel - Keyword-based semantic mapping
🌓 Dark / Light / Auto Theme
⌨️ Keyboard Shortcuts
🔒 Session Persistence - Queue survives refresh
📱 Fully Responsive

🚀 Quick Start

Option 1 - Use the Live Demo (Simplest)

👉 https://tekideki.github.io/AI-Prompt-Builder/

Nothing to install. Just open and start building prompts.

Option 2 - Clone & Run Locally

git clone https://github.com/tekideki/AI-Prompt-Builder.git
cd AI-Prompt-Builder

Then either:

a) Double-click index.html - works, but clipboard / localStorage may be restricted in Chrome/Edge under file://.

b) Run a local server (recommended for full functionality):

# Pick any static server
python -m http.server 8000
# or
npx serve
# or
php -S localhost:8000

Then open http://localhost:8000

Option 3 - Fork & Deploy to Your Own GitHub Pages

Fork this repository
Go to Settings → Pages
Source: Deploy from a branch → Branch: main → Folder: / (root)
Save → your app will be live at https://YOUR-USERNAME.github.io/AI-Prompt-Builder/

🎮 Usage

Basic Workflow

Enter a subject - e.g., a young woman in a red dress
Pick a preset or configure manually:
- 🎭 Visual Style
- 📐 Composition
- 💡 Lighting (source · direction · quality · temperature)
- 📷 Camera (focal length · aperture · motion)
- 🎞️ Film Stock & Quality
- 🎨 Color Palette
- ⚙️ Render Tech & Environment Interaction
- 🌍 Background
Choose your engine in the 🎯 Target AI Engine dropdown
Click 🚀 Generate Prompt (or press Ctrl+Enter)
Copy and paste into your AI tool

Keyboard Shortcuts

Shortcut	Action
`Ctrl` + `S`	Save settings
`Ctrl` + `Enter`	Generate prompt
`Ctrl` + `Shift` + `R`	Smart Randomize
`Ctrl` + `E`	Export SD batch
`Ctrl` + `H`	Toggle history
`Ctrl` + `K`	Focus AI input
`Esc`	Blur active field

On macOS, use Cmd instead of Ctrl. Auto-detected.

Batch Processing

Create a .txt file with one subject per line:

a young woman in a red dress
an old fisherman on a boat
a cyberpunk samurai
a medieval castle at sunset

Drag & drop it onto the upload zone, then use ⏮️ Prev / ⏭️ Next to navigate. Click 💾 Export SD Batch to download all generated prompts.

🎯 Supported Engines

Engine	Syntax Style	Special Parameters
Midjourney v6/v7	Parametric + Discord flags	`--ar`, `--stylize`, `--chaos`, `--weird`, `--style raw`, `--no`
FLUX.1	Natural language	`guidance`, `steps`, `seed`, `aspect`
Stable Diffusion	Tag-based with weights	`CFG`, `Steps`, `Sampler`, `Hires Fix`, `Refiner`, `Face Restoration`
DALL-E 3	Descriptive prose	`Vivid` / `Natural`, `HD` boost

📦 Stable Diffusion Models Included

Click to expand the full model list (18 models)

SDXL 1.0 (Recommended)

Model	Category	Best For
Juggernaut XL v9	Human Realism	General realistic photography
CyberRealistic XL	Human Realism	Portraits, skin detail
AbsoluteReality	Human Realism	Natural colors, female portraits
OdysseyXL-MK2-Alpha	Landscape	Epic landscapes, architecture
Illustrious Anime v4	Anime	Modern SDXL anime
Pony Diffusion V6 XL	Anime	Anime (requires trigger words)
DreamShaper XL Lightning	Turbo	Ultra-fast (6 steps)

SDXL Turbo

Model	Category	Best For
SDXL Lightning	Turbo	4-step generation
Realistic Vision Lightning	Turbo	Fast realism, batches

SD 3.5

Model	Category	Best For
SD 3.5 Large	General	Latest Stability model
SD 3.5 Large Turbo	Turbo	4-step version
SD 3.5 Medium	General	Limited VRAM (8GB)

SD 1.5 / SD 2.1 (Legacy)

Model	Category	Best For
Realistic Vision v5.1	Human Realism	SD 1.5 classic
DreamShaper 8	General	Versatile
Rev Animated	General	Fantasy, illustration
Poralus-Image-1357	Landscape	SD 1.5 landscapes
SD 2.1 768-v	General	Official Stability model
SD 2.1 Base	General	Faster than 768-v

Each model entry includes architecture compatibility, trigger words, file size, native resolution, recommended CFG/steps/sampler, and VAE requirements.

Add your own custom models via the ➕ Add custom model button.

🏗️ Architecture

Single-File Design

index.html  (~2,500 lines, zero dependencies)
├── <style>    → ~700 lines CSS (theme via custom properties)
├── <body>     → ~450 lines HTML (static UI)
└── <script>   → ~1,350 lines vanilla JS
    ├── State management
    ├── Storage (localStorage + sessionStorage)
    ├── Theme management
    ├── History
    ├── Custom presets & models
    ├── MODEL_DATABASE       (18 SD models)
    ├── ENGINE_CONFIGS       (4 engines)
    ├── PRESETS              (14 built-in)
    ├── Prompt builders      (per engine)
    ├── Score calculator     (weighted)
    ├── Conflict detector    (11 rules)
    ├── Batch processors
    ├── Import / Export
    └── Event handlers

Design Principles

Zero dependencies - no frameworks, no build step, no network calls
Config-driven - Add a new engine by adding one ENGINE_CONFIG + one ENGINE_BADGES entry + one buildPromptString case
Event delegation - One listener per container, not per element
Debounced writes - localStorage writes debounced to 400 ms
Graceful degradation - Fallbacks for clipboard, file:// protocol, missing VAE

Local Storage Keys

Key	Scope	Purpose
`aiPromptBuilder_v22`	`localStorage`	Main settings
`aiPromptBuilder_history`	`localStorage`	Prompt history (max 20)
`aiPromptBuilder_theme`	`localStorage`	Theme preference
`aiPromptBuilder_customPresets`	`localStorage`	User presets
`aiPromptBuilder_customModels`	`localStorage`	User SD models
`aiPromptBuilder_subjectQueue`	`sessionStorage`	Current batch queue
`aiPromptBuilder_subjectQueueIndex`	`sessionStorage`	Queue position

The JSON backup contains settings, history, custom presets, custom models and the theme. The subject queue is intentionally session-only and is therefore not part of the backup.

🌐 Browser Support

Browser	Status	Notes
Chrome / Edge 90+	✅	Recommended
Firefox 88+	✅	Best `file://` support
Safari 15+	✅	Minor CSS differences
Brave / Vivaldi / Opera	✅	Chromium-based
Mobile Chrome / Safari	✅	Responsive layout

Live version tested on: tekideki.github.io/AI-Prompt-Builder ✅

`file://` Limitations

When opening the file directly:

⚠️ localStorage may be blocked → falls back to in-memory (session only)
⚠️ navigator.clipboard may fail → automatic document.execCommand fallback
⚠️ sessionStorage works but doesn't persist across tabs

Recommendation: Use the live demo or a local HTTP server.

🐛 Troubleshooting

Prompt doesn't update when I change a dropdown

Check the browser console for errors. Try a hard refresh (Ctrl+Shift+R).

Score doesn't change when I switch options

The score varies based on prompt length and completeness, not on the semantic quality of each choice. Two options of similar length produce the same score. This is by design - see calculatePromptScore().

Copy button doesn't work

If on file://, the app falls back to document.execCommand('copy'). If that also fails, check your browser's clipboard permissions at chrome://settings/content/clipboard.

Import JSON shows "Missing or invalid settings field"

The file isn't a valid Prompt Builder export. Verify it's the JSON downloaded from this tool, not manually edited.

Settings lost after refresh

You're on file:// with localStorage disabled. Use the live demo or switch to Firefox.

Import says "nothing was changed"

The file was rejected before anything was applied, so your current settings, history and presets are untouched. This happens when the JSON is not a Prompt Builder export, when the settings are not in the expected shape, or when a setting value no longer exists in the current version. Use the file produced by Export JSON without editing it.

If some settings are reported as not applied, the backup was written by a different version of the app: the values that are still valid were imported, and the toast tells you how many.

Save says storage is unavailable

The browser refused to store the settings (localStorage blocked, private mode, or a full quota). The app keeps working in this tab, but nothing is persisted until storage is available again. Try the live demo in a regular window, or free some space.

Queue disappeared after closing the tab

By design - the queue uses sessionStorage, cleared when the tab closes. To persist across sessions, change SUBJECT_QUEUE_KEY to use localStorage in the code.

🗺️ Roadmap

🤝 Contributing

Contributions are welcome!

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Commit your changes (git commit -m "Add amazing feature")
Push to the branch (git push origin feature/amazing-feature)
Open a Pull Request

Code style

No dependencies - keep it vanilla
Single file - everything stays in index.html
Comment non-obvious logic
Test on Firefox + Chrome before submitting

📝 Changelog

v22.13 - Current

Security

Custom model names and architecture values are rendered as text, never as HTML (fixes a stored DOM XSS path reachable through an imported JSON backup)
Import now validates the whole file first and rolls back cleanly if anything fails

Fixes

JSON backup restores engine parameters correctly (architecture, checkpoint, CFG, steps, sampler, Hires/Refiner/Face Restoration and the other engine fields)
Switching engines and Batch All Engines no longer overwrite your current parameters
Checkbox settings (Style Raw, Tile, Hires Fix, Refiner, Face Restoration) survive a panel rebuild
Cross-tab sync rebuilds the engine panel instead of showing another engine's controls
History restore, batch copy and Copy All report copy failures honestly instead of a false success
Saving reports a real error when the browser refuses the write, and a blocked localStorage no longer prevents the app from starting
Malformed stored history can no longer break startup
Numeric engine parameters are clamped to their allowed range before they reach the prompt

Accessibility

Every select, number and text field has an associated label
Contrast raised for buttons, hints, labels, score badges, history header and footer in both themes

Improvements

Service worker: offline navigation fallback, no caching of JSON or cross-origin requests, update notification
Added description, canonical and social preview metadata, plus richer web app manifest
Copy/import failures now report the real outcome instead of a success message

v22.12

Fixes

copyPrompt() no longer relies on window.event - works on Firefox now
applyRecommended() no longer pollutes prompt history
No more duplicate event listeners on sdModel / sdVersion (event delegation)
Debounced generatePrompt (200 ms) for smoother typing
Replaced deprecated navigator.platform with detectMacPlatform()
Strict JSON import validation (anti prototype-pollution)
Subject queue persists via sessionStorage + Clear button
Live prompt score with weighted calculation and visual feedback
Unified applyPreset / applyCustomPreset code paths

Improvements

Inline SVG favicon (no more 404)
Score animation (green/red flash on change)
Combo bonuses in scoring (realism, technical, pictorial, light stack)
Score penalty tuned (errors: −10, warnings: −4)

v22.8 - Initial Release

Multi-engine support (MJ, FLUX, SD, DALL-E 3)
14 presets, 18 SD models, batch processing

📜 License

Distributed under the Apache License 2.0. See LICENSE and NOTICE for details.

🙏 Credits

Original concept & UI: @tekideki Model database: Community-curated info from Civitai and Hugging Face

Special thanks to the AI art community for ongoing feedback.

🔗 Resources

⭐ If this tool helped you, please give it a star!

Built with ❤️ for the AI art community

⬆ Back to Top

Install app

Use the Install app button near the footer. On supported browsers it opens the native installation prompt when available. Otherwise it shows localized instructions for Safari on iPhone/iPad, Android, and desktop browsers. Installation is controlled by the browser; no installation is claimed without its confirmation. The button is hidden in a standalone/minimal-ui app window and after an appinstalled event in the current session. A normal browser tab cannot always detect a previous installation.

Installation needs HTTPS or trustworthy localhost. The controls add no dependencies, tracking, network calls or storage keys; existing application data is not modified. On an existing cached version, revisit online and refresh after the service worker has updated to see the button. No automatic reload is forced.

Name		Name	Last commit message	Last commit date
.github		.github
LICENSE		LICENSE
NOTICE		NOTICE
README.md		README.md
apple-touch-icon.png		apple-touch-icon.png
favicon.ico		favicon.ico
icon-192.png		icon-192.png
icon-512.png		icon-512.png
icon-maskable-512.png		icon-maskable-512.png
index.html		index.html
manifest.json		manifest.json
service-worker.js		service-worker.js

Latest commit

History

Folders and files

Repository files navigation