Vertical Video Converter

Automatically reframes landscape video (e.g., 16:9) to vertical format (9:16) using computer vision to detect and track subjects/action within the frame, applying intelligent cropping that keeps the primary subject centered rather than naive pillarboxing. The system analyzes frame content across the video timeline to maintain temporal consistency during the crop operation, though the specific vision model architecture (CNN, transformer, optical flow) and training approach remain undocumented.

Adds a blurred background to the sides of a landscape video when converting to vertical format, preserving the full original content without cropping. The system analyzes the source video's color palette and applies a blur filter to the extended background, maintaining visual coherence between the original content and the added fill area. This approach avoids information loss from cropping but increases file size and may distract from the primary subject.

Implements a freemium SaaS model where users can perform one free 60-second conversion without signup, then must provide email and upgrade to paid tier for additional conversions. The system enforces quota limits at the application level: free tier allows unlimited single conversions but only one per user (tracked via browser/IP), while paid tier ($10/month) allocates 60 minutes of total processing time per month. Quota tracking and enforcement happen server-side after file upload and processing completion.

Accepts video file uploads via web form (max 250MB free tier, 1GB paid tier), processes the file on remote servers using undocumented infrastructure, and returns a downloadable vertical video file. The system does not support real-time preview, batch processing, or API access — all interaction happens through the web UI. Processing latency, output codec, and bitrate are not documented, making it impossible to assess quality or performance characteristics.

Claims to detect and track 'action' and subjects within video frames to inform intelligent cropping decisions, keeping primary subjects centered during the landscape-to-vertical conversion. However, the specific detection mechanism (object detection model, saliency maps, optical flow, face detection) is proprietary and not disclosed. The system appears to analyze multiple frames to maintain temporal consistency, but the algorithm and confidence thresholds are unknown. Accuracy and failure modes are not documented.

Implements server-side quota tracking that allocates 60 minutes of video processing per month for paid tier users ($10/month), enforced at the application level after file upload and processing completion. Quota resets on a calendar month basis (specific reset time undocumented). Once monthly quota is exhausted, further conversions are blocked until the next month or user upgrades to enterprise tier. No overage pricing, burst capacity, or quota rollover is available.