A lightweight browser-based AI system for personalized pixel drawing.
The system learns from your drawing habits in real-time and assists you by predicting your next strokes using frequency maps combined with a small Convolutional Neural Network (CNN).
- User-Adaptive AI: Trains directly from your inputs for personalized drawing predictions.
- Frequency Maps: Tracks color, position, and neighboring pixels to enhance prediction quality.
- High-Precision Learning: Values stored numerically (0–10,000) for more accurate percentages during calculations.
- Two AI Modes:
- Fast: Uses only frequency maps and minimal CNN for real-time assistance.
- Accurate: Uses all frequency maps and full CNN for higher quality, slower predictions.
- Canvas Customization: Adjustable canvas size and brush pixel size.
- Coordinate Drawing: Draw pixels precisely via input coordinates.
- Upload Images: Train AI using your drawings or existing images.
- Save & Load Models: Store AI models and frequency maps locally.
- Cache & Preferences: Stores user settings locally; includes a clear-cache option.
- Mobile Friendly: Works on touch devices, though desktop use is recommended for convenient experimentation.
- Clone or download the repository.
git clone https://github.com/YourUsername/AI-Assisted_Pixel_Drawing.git
- Open index.html in your browser. No server required — everything runs locally.
Usage Tips
Recommended for Desktop: Easier for precise input and full-feature experimentation.
AI Modes:
Fast Mode: Quick predictions for real-time assistance.
Accurate Mode: Higher-quality AI-assisted drawings at the cost of speed.
Brush Size: Adjust for finer or broader strokes.
Coordinate Input: Draw exact pixels; the system prevents out-of-canvas errors.
Saving Work: Save your frequency maps and models for future training.
Technical Notes
Runs entirely in-browser using TensorFlow.js.
Frequency maps store color, position, and neighbor frequency numerically for high precision.
Training and prediction are fully local; no data leaves your browser.
Supports multiple image uploads for batch training.
Contributing
Contributions and feature suggestions are welcome! Please open an issue or submit a pull request.
License
MIT License
Originality Note
This system is an original design by the author. It combines frequency-based pixel learning, precision scaling, and CNN-assisted prediction for a personalized, local AI-assisted drawing experience. Coordinate Drawing: Input specific coordinates to place pixels precisely.
Upload Images: Train AI using existing drawings or images.
Save & Load: Save AI model and frequency maps locally.
Cache & Preferences: Stores user settings locally; includes a clear-cache option.
Mobile Friendly: Works on touch devices (though desktop use is recommended for convenient experimentation).
Installation
Open index.html in your browser. No server required — everything runs locally.
Usage Tips
Recommended for Desktop: While the app works on mobile, experimenting and training is more convenient on a desktop due to larger screen and precise input.
AI Modes:
Use Fast mode for real-time assistance.
Use Accurate mode when you want higher-quality AI-assisted drawings, even if slower.
Pixel Size: Adjust the brush size for finer or broader strokes.
Coordinate Input: Use the coordinate inputs to draw exact pixels. The system prevents out-of-canvas errors.
Saving Work: Always save your frequency maps or model if you want to continue training later.
Technical Notes
Runs entirely in-browser using TensorFlow.js.
Frequency maps store color, position, and neighbor frequency numerically for high precision.
Training and prediction are local; no data is sent to a server.
Supports multiple image uploads for batch training.
Contributing
Contributions and feature suggestions are welcome! Please open an issue or submit a pull request.
License
MIT License