Flask backend for CipherMuse - Advanced steganography platform with AI generation and user authentication.

• User registration and login
• Password hashing with SHA-256
• Session management
• User-specific data storage

• Image Generation: Stable Diffusion v1.5 with MSE-840000 VAE
• Music Generation: Facebook's MusicGen-Small model
• Image Upscaling: SwinIR for enhanced quality (optional)
• Halloween Themes: Spooky, Gothic, Witch, and Zombie styles

• LSB Matching: Least Significant Bit steganography with PRNG positioning
• AES-256 Encryption: Password-based encryption with SHA-256 hashing
• Multi-format Support: Images (PNG, JPG, BMP) and Audio (WAV, MP3, M4A)

• User-specific file galleries
• Encrypted file storage and retrieval
• Download and preview functionality

1. Install dependencies:

pip install -r requirements.txt

2. Download and place AI models in the models directory:

models/
├── v1-5-pruned-emaonly-fp16.safetensors    # Stable Diffusion v1.5
├── vae-ft-mse-840000-ema-pruned.safetensors # MSE VAE
└── SwinIR.onnx                              # SwinIR upscaler (optional)

Model Sources:

• Stable Diffusion v1.5: Hugging Face
• MSE VAE: Hugging Face
• SwinIR: Official Repository
• MusicGen: Auto-downloaded from Hugging Face

3. Run the server:

python app.py

Server starts on http://localhost:3001

• POST /api/register - Create new user account
• POST /api/login - User login

• GET /api/styles - Get available Halloween themes
• POST /api/generate-image - Generate Halloween images
• POST /api/generate-music - Generate spooky music

• POST /api/encrypt-image - Hide message in AI-generated image
• POST /api/decrypt-image - Extract message from image
• POST /api/encrypt-audio - Hide message in AI-generated music
• POST /api/decrypt-audio - Extract message from audio

• GET /api/gallery/<username> - Get user's gallery
• POST /api/gallery/<username> - Save gallery data
• GET /api/download/<filename> - Download encrypted files

• Python 3.8+
• CUDA-capable GPU (recommended, 6GB+ VRAM)
• 16GB+ RAM (8GB minimum)
• 20GB+ disk space for models and temporary files

• Stable Diffusion v1.5: RunwayML/Stability AI
• MusicGen-Small: Meta AI (Facebook Research)
• SwinIR: Jingyun Liang et al. (ETH Zurich)
• MSE VAE: Stability AI

• SHA-256 password hashing
• AES-256 message encryption
• LSB steganography with pseudo-random positioning
• User-isolated data storage
• Secure temporary file handling
• CORS protection

This project uses several open-source models:

• Stable Diffusion v1.5: CreativeML Open RAIL-M License
• MusicGen: MIT License
• SwinIR: Apache License 2.0

Please ensure compliance with respective model licenses when using this software.