Update app/database.py

Update app/database.py

Update app/database.py

Add ai_edit_daily_count tracking to media_clicks collection

Change ai_edit_last_date to MongoDB Date object format - Store as Date object instead of string format

Add ai_edit_complete and ai_edit_last_date to media_clicks collection

Add success/failure logging to colorizations collection

Only log to media_clicks when user_id is provided

Update default fallback category ID to 69368fcd2e46bd68ae1889b2

Set admin database name to adminPanel for media_clicks collection

Fix empty string handling and support both subCategories and categories schemas

Add MongoDB media click logging with per-user category counts

Fix MongoDB database boolean check - use 'is None' instead of truthiness

Fix Gradio compatibility: remove show_copy_button parameter

Add MongoDB integration for API call logging with timestamps

Update Firebase initialization to auto-detect credentials file and add setup guide

Add Firebase email/password authentication endpoints - Add /auth/register endpoint for user registration - Add /auth/login endpoint with Firebase REST API integration - Add /auth/me endpoint to get current user info - Add /auth/refresh endpoint for token refresh - Integrate with existing Firebase App Check - Add httpx and email-validator dependencies

Refactor to use Hugging Face Inference API with fal-ai provider - Replace local model loading with InferenceClient API - Remove heavy SDXL/ControlNet/BLIP model dependencies - Use FLUX.1-Kontext-dev model via API - Keep FastAPI and Firebase authentication - Significantly reduce memory usage (no local models)

Add memory optimizations for SDXL pipeline - Enable VAE slicing and tiling - Enable attention slicing for UNet and ControlNet - Use sequential CPU offloading for pipeline - Keep BLIP model on CPU to save GPU memory - Add torch.no_grad() and cache clearing - Reduce guidance scale for lower memory usage

Fix permission errors and environment variables for model loading - Use /tmp directory for writable model storage - Set OMP_NUM_THREADS and cache directories at startup - Add better error handling and permission checks

Add FastAPI version using SDXL + ControlNet for text-guided colorization (from fffiloni)

Improve health check and UI messaging for fallback mode

Improve fallback colorization with better color hints and saturation

Add simple colorization fallback using LAB color space when model fails to load

Update app/main.py

Update app/main.py

Clean up ResNetGenerator architecture - simplified structure

Fix: Try ResNet first, improve size validation, and skip configs with too many mismatches

Improve weight loading verification and skip configs with too many missing keys

Fix colorization: Add ResNet generator architecture and fix minimum image size to prevent kernel errors

Improve PyTorch colorizer: Better preprocessing, output handling, and debugging

Fix image colorization: Add PyTorch GAN colorizer fallback, update Dockerfile to use main_fastai, and add missing dependencies

Update app/main.py

Update app/main.py

Update app/main.py

Update app/main.py

Update app/main.py

Update app/main.py

Update app/main.py

Add guide for finding FastAI-compatible colorization models - Document how to find models with .pkl files - Add instructions for setting MODEL_ID - Note current model limitation

Fix Gradio permission error by setting GRADIO_TEMP_DIR to /tmp

Add fallback logic to handle PyTorch models and find actual model files - Add manual download fallback when from_pretrained_fastai fails - List repository files to find .pkl or .pt files - Provide clear error message for PyTorch models

Integrate FastAI colorization with Firebase auth and Gradio UI - Replace main.py with FastAI implementation - Add Gradio interface for Space UI - Add Firebase authentication to /colorize endpoint - Add curl examples documentation - Update test.py with User-Agent headers

Detect PyTorch models and provide clear error message - Repository contains generator.pt (PyTorch) not FastAI model

List repository files to find actual model filename instead of guessing

Add fallback mechanism to manually download and load FastAI model if from_pretrained_fastai fails

Fix cache directory permissions and OMP_NUM_THREADS warnings - Set HF cache env vars before imports - Set MPLCONFIGDIR for matplotlib - Fix OMP_NUM_THREADS in Dockerfile

Switch to FastAI GAN colorization model (Hammad712/GAN-Colorization-Model)

Restore local ControlNet colorization pipeline

Point inference requests to new HF router endpoint