feat: implement premium cinematic typography layouts, revert ControlNet, and remove token

feat: Enable mock mode by default and add comprehensive generation logging Frontend Changes: - Change mock mode default from false to true for faster testing - Users can toggle to real mode via UI toggle switch Backend Changes: - Add detailed logging for model loading stages (download, move to device, etc.) - Add warning message about large model download (4GB+, 5-10 minutes) - Add timeout handling for model loading - Add detailed inference logging (steps, guidance_scale, seed) - Add generation completion summary logging - Better error messages guiding users to use mock mode or wait for download This fixes the stuck generation issue by: 1. Defaulting to mock mode which generates instantly (~1-2 seconds) 2. Providing clear logging when downloading real model 3. Gracefully falling back to mock if real model fails to load

fix: Improve Ollama client logging and use llama3.2:1b model - Add debug logging on OllamaClient initialization - Update model to llama3.2:1b (faster, lighter weight) - Improve connection check logging with clear status messages - Graceful fallback to heuristic prompts when Ollama unavailable Ollama is running on localhost:11434 with multiple models available: - llama3.2:1b (1.2B, fastest) - llama3:latest (8B, best quality) - mistral:latest (7.2B, balanced) - deepseek-r1:7b (7.6B, best reasoning)

fix: Improve Ollama fallback handling and add better debug logging - Reduce Ollama timeout from 35s to 10s for faster fallback - Add TimeoutError handling alongside URLError - Improve debug messages for model loading stages - Enable mock mode by default (can be toggled via API) - Add more granular logging for MPS pipeline initialization This fixes the hanging issue when Ollama is not running by quickly falling back to mock mode with better error reporting.

v1.1: Add 16 specialized categories, colorization, face restoration, and advanced image editing Features: - Add category_prompts.py with 16 specialized categories (110+ templates) * Creative Art, Characters, Landscapes, Architecture, Vehicles, Products, Marketing * Food, Fashion, Gaming, Animals, Events, Business, Education * Each category includes 5-8 optimized subcategories with specialized prompt templates - Implement colorization endpoint (/api/colorize) with 5 styles * Vibrant, Warm, Cool, Vintage, Sepia color grading styles * Adaptive color enhancement with saturation and contrast control - Implement face restoration endpoint (/api/face-restoration) with 4 intensity levels * Low, Medium, High, Ultra intensity levels * Denoising, sharpening, contrast enhancement, color vibrancy boost - Update pipeline.py with new methods * colorize() method supporting 5 color grading styles * restore_face() method with intensity-based enhancement - Update app.py with new request models * ColorizeRequest model with style parameter * FaceRestorationRequest model with intensity parameter * Add 2 new endpoints: /api/colorize and /api/face-restoration - Update ollama_client.py expand_prompt() method * Add category and subcategory parameter support * Integrate category-specific prompt enhancements * Maintain backward compatibility Backend: - All endpoints tested and verified - Proper error handling and base64 encoding/decoding - Rate limiting applied to new endpoints - Production-ready implementation

Deploy LumaForge AuraGen backend API to Hugging Face