A powerful, professional-grade image generation tool that runs completely offline on your own hardware. Generate unlimited images with zero costs, no paid subscriptions, no per-image charges, and complete privacy.
⚠️ Setup Note: Requires free HuggingFace account for model downloads. No ongoing API costs or per-image charges.
- Installation Guide - System requirements, dependencies, and setup
- Models & Comparison - Complete model reference with visual quality comparisons
- Usage Guide - How to use the tool, examples, and tips
- Troubleshooting - Common issues and solutions
# Simple generation (uses simple_prompts.json)
python generate_images.py simple_prompts.json output/
# With custom aspect ratios (uses sample_poster_prompts.json)
python generate_images.py sample_poster_prompts.json output/
# Compare multiple models
python generate_images.py sample_poster_prompts.json output/ multiFirst time? Check the Installation Guide for setup instructions.
Example Files Included:
simple_prompts.json- 5 detailed prompts optimized for quality (sunset, forest, ocean, city, flowers)sample_poster_prompts.json- Same 5 prompts with custom aspect ratios (square, portrait, landscape, portrait wide, cinematic)
Both files use detailed, descriptive prompts (~77 tokens) for better image quality.
- No paid API services - runs 100% locally
- Zero per-image costs
- No rate limits
- All models are open-source
- Works completely offline after model download
- Your prompts never leave your computer
- No data collection or tracking
From lightweight (4GB) to ultra-quality (34GB). Each model specializes in different strengths - photorealism, artistic style, speed, or maximum quality. See Models & Comparison for visual examples.
This tool goes far beyond simple single-image generation:
- One idea → Multiple variations: Express concepts through different prompt perspectives
- Batch everything: Process multiple ideas × multiple prompts × multiple models × multiple aspect ratios - all in one execution
- Compare models instantly: Run the same prompts through all 7 models simultaneously to find the perfect match
Square, Portrait, Landscape, Cinematic, and more. Per-image resolution control via JSON.
No sequential processing - everything runs in parallel. Generate dozens of variations while you grab coffee.
| Feature | This Tool | MidJourney | DALL-E 3 |
|---|---|---|---|
| Cost | FREE | $10-60/mo | $0.04/img |
| Privacy | 100% Private | Public | Stored |
| Offline | ✅ Yes | ❌ No | ❌ No |
| Batch Mode | ✅ Unlimited | ❌ Limited | ❌ No |
| Multi-Model | ✅ 7 models | ❌ 1 model | ❌ 1 model |
- Content Creators: Unlimited thumbnails, banners, social media graphics
- Designers: Mockups, concepts, and variations without stock photo costs
- Developers: App icons, food images, UI elements for projects
- Marketers: Product visuals, ads, promotional materials at zero cost
docs/images/
├── generate_images.py # Main script
├── README.md # This file (overview)
├── INSTALLATION.md # Setup guide
├── MODELS.md # Model reference
├── USAGE.md # Usage examples
├── TROUBLESHOOTING.md # Common issues
├── simple_prompts.json # Simple example (5 basic prompts)
├── sample_poster_prompts.json # Advanced example (custom aspect ratios)
└── assets/[output folders]/ # Generated images
- ✅ Fixed MPS NaN issue - Changed to float32 on Apple Silicon to prevent black images
- ✅ Removed unnecessary defensive logic - Cleaner, simpler code
- ✅ Fixed safety_checker warnings - Proper None assignment
- ✨ Added custom aspect ratio control (6 presets)
- ✨ Per-prompt resolution control via JSON
- ✨ Three resolution modes (JSON, Override, Default)
- 📝 Split documentation into focused files (INSTALLATION, MODELS, USAGE, TROUBLESHOOTING)
⚠️ Marked Chroma-GGUF (Model 8) as broken due to GGUF loader incompatibility
⚠️ Attempted GGUF loading (unsuccessful)- Added Chroma-GGUF model support (now broken - awaiting fix)
- Added FLUX.1-schnell support
- Multi-model comparison mode
- Cross-platform device detection
Found a better model? We welcome improvements! See our Models & Comparison Guide for:
- How to test new models with
sample_poster_prompts.json - PR requirements (code, outputs, metrics, documentation)
- Performance benchmarking guidelines
Help the community discover better models for image generation! 🎨
MIT License - Free for personal and commercial use
- Black Forest Labs (FLUX.1-schnell)
- Stability AI (Stable Diffusion)
- HuggingFace (Model hosting & diffusers library)
- All model creators and the open-source AI community