The most Purr-fect Image File Format for your AI workflows
|
Well- yes, but no, I'll come to that bit later, but before that, let me geek out about what this project is MEOW (Metadata Encoded Optimized Webfile) is a Python-based image file format designed to be efficient, practical, and cross-platform compatible. With support for RGBA transparency, metadata, and fast rendering capabilities, MEOW provides a modern alternative for image storage and manipulation. Whether you're a developer looking for a lightweight image format, an AI locked-in person to use the best possible file fomat, a digital artist needing transparent image support, or just a curious coder who wants a .meow file extension, MEOW offers a simple yet powerful solution. |
The First image to ever be converted to .meow Note: GitHub doesn't support .meow files (YET), so had to display the original PNG source |
MEOW started off as just a fun, new (very ineffecient) image file format I made studying about them as a meme while having better metadata to be used with AI, but I realised something.
People don't make new file formats simply because of one problem- they're very hard to bring to mass adoption, what if we fix that?
MEOW uses a steganographic image format that embeds metadata that's needed for AI workflows while maintaining PNG compatibility. It's basically PNGs on Steroids.
|
Unlike traditional formats that require special viewers, MEOW files contain standard PNG data:
Setup Required: Either rename |
|
MEOW is specifically designed for use with AI, embedding metadata that is really useful for machine learning workflows and potentially can enhance LLM image understanding
|
|
MEOW embeds AI-specific parameters that traditional formats can't (or well, can with a ton of modification can) handle. When any image is converted to a .meow, this data is automatically added in the conversion process:
- Pre-computed AI Features
- Optimal preprocessing parameters embedded in the file
- Attention maps showing where AI models should focus
- Bounding boxes and object detection data
- Saliency regions for computer vision tasks
-
Model Optimization Data
-
Steganographic Storage
Unlike traditional metadata that's easily lost:
- Encoded in pixel LSBs (Least Significant Bits)
- Survives file operations that would strip normal metadata
- Invisible to standard viewers but accessible to AI applications
MEOW uses LSB (Least Significant Bit) steganography to hide AI data inside standard PNG images:
|
Data Structure: |
|
|
| Feature | MEOW Steganographic | PNG | Custom AI Formats | Traditional MEOW |
|---|---|---|---|---|
| Universal Viewer Support | ✅ After setup (rename/association) | ✅ | ❌ Requires special software | ❌ Requires MEOW viewer |
| AI Metadata | ✅ Rich & hidden | ❌ | ✅ But not cross-compatible | ✅ But not cross-compatible |
| File Extension Flexibility | ✅ .meow or .png | ✅ | ❌ Proprietary extensions | ❌ .meow only |
| Cross-Platform | ✅ Cross-platform (with setup) | ✅ | ❌ Limited support | ✅ |
| AI Training Ready | ✅ Embedded annotations | ❌ | ✅ But compatibility issues | ✅ But compatibility issues |
| Data Integrity | ✅ Survives file operations | ✅ | ❌ Often lost in transfers | ❌ Lost if opened wrong |
The Problem: AI-enhanced formats typically sacrifice compatibility for features
The Solution: Hide AI data inside universally-compatible PNG files
The Result: Best of both worlds
8000
- works everywhere + AI superpowers
⚠️ Important: While MEOW files contain standard PNG data, you need either:
- Rename
.meow→.png(compatibility for viewing), OR- Run file association scripts (one-time setup to make
.meowextension recognized)Without setup, most viewers won't recognize the
.meowextension by default. Because well, I just made it lol.
# Clone the repository
git clone https://github.com/kuberwastaken/meow.git
# Navigate to the directory
cd meow
# Install dependencies
pip install -r requirements.txt
# Set up file associations (optional)
windows\associate_meow.bat # Run as administrator# Convert any image to steganographic MEOW
python meow_format.py image.jpg output.meow
# Convert PNG with rich AI annotations
python convert.py image.png enhanced.meow# Launch the MEOW GUI with AI features
python meow_gui.py# Option 1: Rename to .png (works instantly everywhere)
rename test.meow test.png
start test.png # Opens in default viewer
# Option 2: Set up file associations (makes .meow recognized)
windows\associate_meow.bat # Run as administrator first
start test.meow # Now opens directly!# Launch MEOW viewer to see hidden AI metadata
python meow_gui.py
# Run compatibility demonstration
python final_demonstration.py# 1. Create a steganographic MEOW file
python meow_format.py "photo.jpg" "photo.meow"
# 2. Test universal compatibility - Choose your method:
# Method A: Rename to PNG (instant compatibility)
copy photo.meow photo.png # Keep original + create PNG copy
start photo.png # Opens in any image viewer!
# Method B: File association setup (one-time setup)
windows\associate_meow.bat # Run as admin (one-time setup)
start photo.meow # Now .meow files open directly!
# 3. Extract AI data (MEOW-aware apps see hidden metadata)
python meow_gui.py # Load .meow OR renamed .png - AI data intact!To make .meow files open directly in your system's default image viewer, run the appropriate setup script for your platform:
# Run as administrator to set up file associations
windows\associate_meow.bat# Make executable and run
chmod +x macos/associate_meow_macos.sh
./macos/associate_meow_macos.sh# Universal setup script that detects your OS
chmod +x scripts/associate_meow_crossplatform.sh
./scripts/associate_meow_crossplatform.shAfter running the appropriate script, double-clicking any .meow file will open it in your system's default image viewer (Paint, Preview, etc.) while preserving the hidden AI metadata for MEOW-aware applications.
Note: File association is optional - you can always rename .meow files to .png for instant compatibility with any image viewer.
|
MEOW files ARE valid PNG files with hidden data embedded using LSB steganography:
|
|
{
"version": 2,
"features": {
"brightness": 126.642,
"contrast": 67.335,
"edge_density": 0.738,
"mean_rgb": [126.69, 126.6, 126.64],
"dimensions": [400, 300]
},
"attention_maps": {
"avg_attention": 117.86,
"max_attention": 255,
"attention_peaks": 12,
"focus_regions": [[120, 80], [250, 150]]
},
"ai_annotations": {
"object_classes": ["cat", "background"],
"bounding_boxes": [...],
"preprocessing_params": {
"mean_rgb": [0.485, 0.456, 0.406],
"input_size": [224, 224],
"normalization": "imagenet"
},
"llm_context": {
"scene_description": "A domestic cat sitting on wooden surface",
"visual_elements": ["furry texture", "natural lighting", "indoor setting"],
"suggested_tags": ["pet", "animal", "indoor", "portrait"]
}
},
"model_hints": {
"recommended_models": ["ResNet50", "CLIP", "YOLO"],
"complexity_score": 0.73,
"processing_priority": "high_detail"
}
}|
Complete interface with AI metadata viewer, steganographic converter, and cross-compatibility testing |
Command-line tool for embedding AI data in PNG-compatible MEOW files |
PNG-compatible format that works in any viewer after simple setup |
- Built with Python 3.6+
- Uses Pillow/PIL for image processing
- NumPy for efficient steganographic operations
- zlib for AI metadata compression
- tkinter for cross-platform GUI
- JSON for structured AI data storage
| Metric | Standard PNG | Steganographic MEOW | Difference |
|---|---|---|---|
| Viewer compatibility | 100% | 100% (after setup) | Simple setup required |
| Visual quality | Perfect | Perfect | Imperceptible |
| AI data capacity | 0 bytes | 650+ bytes | Rich metadata |
| Load time | Fast | Fast | No noticeable difference |
| LLM context understanding | Basic | Better | Significantly improved |
- AI Training Datasets: Embedded annotations eliminate separate metadata files
- Computer Vision: Pre-computed features accelerate model inference
- Digital Asset Management: Rich metadata without database dependency
- Research Archives: Self-documenting images with analysis results
- Production Workflows: Integration with existing tools (after setup or rename to .png)
- LLM Vision Tasks: Enhanced multimodal AI with embedded context
- Automated Content Analysis: Self-describing images for content pipelines
- AI Model Training: Consistent, portable annotations across platforms
Contributions are welcome! Please feel free to submit a Pull Request.
- Fork the repository
- Create your feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add some amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
Initial idea inspired by FaceDev, whose BRUHIFF format provided a creative starting point.
This project is released under the Apache 2.0 License. See the LICENSE file for details.
Made with ❤️ by Kuber Mehta
Purr-fectly optimized (I mean- as far as my ugly code takes it)