Fixed the geometry perspective to follow the article from Farid.

Fixed visual problem of the vanishing heatmap. Also fixed redundant logs

Fixed Mobile support. Added a new AI media detection mechanism by Farid, that creates geometric lines. Fixed logs

Fix: delete file FK violation (delete feedback first), fix sound autoplay policy

Fix: switch email to Brevo HTTP API (bypasses HF SMTP port blocking)

Switch email to Brevo, fix SMTP config

Fix: 404 page loader stuck, navbar loading spinners, bulletproof NR import

Added email code verif

Fix login (audit_logs table + fault-tolerant audit), feedback graceful errors, remove raw AI explanation fallback, localStorage feedback persistence, Betterstack cloud logging

Fixed loader stuck, Added a new feedback mechanism, and fixed sound upon analysis

Fix model auto-generation, sync cinematic loaders, update guest CTA

Fix Google OAuth hardcoded localhost redirect

Fix Celery Upstash Redis SSL compatibility

Fix database URL driver normalization for psycopg2

Fix Windows-specific dependencies for Linux Docker build

Switch to Hugging Face Spaces Docker deployment

Fix Render pricing: Combine web and worker into single free instance

[Vercel Deployment]

[Deployment Phase 1 & 2] - Neon DB and R2 with working turnstile in login page implemented

[UI/UX] Final touches before deployment.

[UI/UX] Added Account deletion, with some new features too.

[UI/UX] More changes to the Frontend

[Minor changes to Frontend]

[UI/UX] Features and good animations added

[Frontend] - Initial Push

Minor changes to message of file_routes, to be more informative.

[Updated Feature] > Updated every file_route, user_route, so that guest users can login and only upload 5 videos, 1 image per IP

[Feature Updated] > Updated anomaly_gif_maker a lot, to be able to precisely extract frames from the video, where the model and pipeline think that part is AI.

[Feature Added] > anomaly_gif_maker. This feature detects the frame where the pipeline thinks it is AI, converts it to gif and shows a side by side comparision to the Original Video, while also annotating the 'Peak Glitch' on the AI RADIANCE MAP gif

[Feature Added] > Added gif maker for video anomaly detector, where the AI detects where the anomaly is and generates a small gif with comparision to Original and AI Radiance Map.

[Feature Added + Updated] > Addediffusion spectrum analyzer which analyzes the video in a more mathematical manner and detects if it's AI or not. Also updated some files to support diffusion spectrum analyzer

[Feature Added] > Model Drift Detection mechanism. This feature checks, if the model is confident in predicting the new uploaded images/videos as AI or not, if not then we need to retrain the model.

[Feature Added] > Added Cloudflare Turnstile captcha for better security and bot disengagement. Also updated features, like file_routes, auth_routes to support Turnstile.

[Feature Updated] > Tweaked aggregator for video processing a bit to identify videos more effectively

[Feature Updated] > Updated response for files such that, the video files only get the confidence and ai_explanation added, while image files do not.

[Feature Added] > Video Pipeline. Have to update the confidence and ai_analysis param for images, where they won't appear.

[Feature Added] > Added the Ensemble Aggregator which acts as the Supreme Court of the Pipeline. Listens to all the four distinct detectors and weighs evidence to a final verdict.

[Feature Added] > Video metadata analyzer. This extracts the metadata from any video and judges it on those information scrapped.

[Updated Features] > Updated explainer and attribution with a new model, and support for better heatmap.

[Feature Added] > motion detector. This detects motion using Farneback Flow, which tracks motion using mathematical equations. If motion is violent and closer to 1.0 - It is AI, If motion is smooth and lower - it is Natural.

[Feature Added] > frame_detector. This detects frames and spits out AI or REAL, based on the provided Video. Tested completely, with EDGE CASES.

Feature Added -> frame_extractor (Extracts Frames safely. Tested with EDGE Cases too, where graceful termination is a must.)

Added Secure Video Upload Validation

Added alembic for graceful migration of database, without creating new columns/altering columns physically. Alembic does it now.

Added separate storage and fixed worker, tasks, which previously got stuck on PROCESSING (NULL PROCESSOR)

Added HTTP Security Headers, for more security

Added exception handler for global exceptions

Added audit logs to be logged into database for staff ease

Initial commit of full project [remaining backend some parts and frontend to be done yet]