fix(analyzer): advanced robust extraction with validation and buffer fallback

fix(analyzer): correction for ffmpeg user_agent flag syntax

fix(analyzer): restore full downloader service with robust TikTok extraction

fix(analyzer): final robust TikTok fix with URL resolution & header standardization

fix(analyzer): restore missing get_direct_url function to fix name error

fix(analyzer): restore missing get_video_info and download_video functions to fix import error

fix(analyzer): implement iterative stream extraction fallback to bypass ip blocks

fix(analyzer): add referer headers and probesize to fix ffmpeg decoder error

fix(analyzer): properly quote ffmpeg scale filter to fix syntax error

fix(analyzer): robust FFmpeg streaming with user-agent and reconnection

fix(analyzer): robust doubled-URL normalization and enhanced error reporting

fix(mobile): optimize frame extraction with scaling and allow direct video URLs

fix: refined URL normalization and mobile TikTok support

fix: update yt-dlp and improve logging

chore: full sync with local backend core and services

Enhance downloader and downloader fixes

fix: implement local metadata extraction

feat: implement SightEngine AI Music detection

fix(pipeline): normalize decorative scores (0-1 range)

fix: critical syntax error in pipeline.py (duplicate except)

fix: obscure provider identity in backend API

feat: add Teacher-Student data collection pipeline

fix: propagate decorative scores to signal objects for frontend display

feat: decorative scoring - SightEngine only, audio/meta derived from visual

Upgrade audio analysis to use HF Inference API

FIX: Add TikWM API for reliable TikTok downloads

FEATURE: YouTube thumbnail fallback when video download is blocked

FIX: YouTube extraction - use ios/web player clients (android blocked)

FIX: YouTube Shorts URL conversion for better yt-dlp support

ENHANCED: 8 metadata signals + 8 heuristic signals for better detection

ENHANCED MODEL: 8 frames, temporal analysis, 2-model ensemble, 6 audio signals

Enhance confidence: 5-factor calculation, HIGH confidence for SightEngine

Deploy VeriVid Backend to Hugging Face Spaces