Face morphing in real-time browser environments demands more than visual fidelity; it requires stability under motion, interpretability of neural outputs, and performance that scales reliably across consumer hardware. This work presents MorphAI v1.5, a production-grade upgrade to the animated face-swapping framework introduced in v1, extending its capabilities with a significantly optimized WebGL rendering pipeline, a re-engineered multi-subject tracking architecture, and a comprehensive suite of diagnostic overlays designed to make facial computation fully observable and verifiable. The system sustains 60 frames per second on modern mobile hardware through deliberate low-level optimizations, including reduced draw calls, efficient buffer management, and intelligent frame synchronization that eliminates jitter and temporal artifacts. Multi-subject detection is handled by a redesigned bounding box system that maintains stable alignment across dynamic scenes involving rapid movement, scale variation, and orientation shifts, ensuring that every morph operation remains spatially coherent and distortion-free. To advance interpretability, MorphAI v1.5 introduces three diagnostic overlays: a high-density Face Mesh that maps a polygonal topology directly onto detected faces for sub-millimeter tracking precision; a 68-point Landmark system that anchors expression mapping and feature alignment to consistent anatomical reference nodes; and a Heatmap overlay that renders temporal velocity gradients across facial regions, surfacing micro-expressions and motion dynamics imperceptible to the human eye. A Split View Engine further enables side-by-side comparison between source input and neural output, transforming the pipeline from an opaque black-box system into a fully auditable visual environment. The framework additionally supports seamless export of morphed outputs beyond the live canvas and integrates a keyboard shortcut system for streamlined control in power-user workflows. MorphAI v1.5 is implemented as a browser-native system optimized for consumer-grade hardware and is publicly deployed on Hugging Face. These advancements collectively establish a robust foundation for interactive media, creative visual tooling, and research applications demanding uncompromising performance, precision, and transparency. Features 1. FPS Tracker (Performance Validation Layer) MorphAI v1.5 introduces a real-time FPS Tracker that functions not merely as a display metric but as an active validation layer for rendering consistency. It continuously monitors frame delivery to confirm that the system sustains 60 frames per second under live morphing conditions. This is made possible through coordinated optimizations across the WebGL pipeline, including reduced draw calls, optimized buffer management, and precise frame synchronization, ensuring that performance remains stable and predictable across a wide range of devices and scene complexities. 2. Multi-Subject Bounding Box System (Stability and Alignment) The bounding box architecture has been fully re-engineered to support reliable multi-subject detection and tracking across dynamic scenes. Each detected face is continuously identified and followed, with adaptive cropping that responds to changes in scale and orientation without disrupting alignment. Advanced stabilization logic handles rapid subject movement, ensuring that morph operations remain spatially consistent and visually coherent regardless of scene complexity or the number of subjects present. 3. Split View Engine (Neural Output Validation) The Split View Engine enables real-time, side-by-side comparison between the original input and the processed neural output. This feature provides developers and researchers with direct visibility into pixel-level transformations, allowing morphing accuracy to be validated frame-by-frame and visual artifacts to be identified and addressed with precision. By making every transformation directly inspectable, Split View converts MorphAI from a black-box inference system into a transparent and fully accountable visual pipeline. 4. Diagnostic Overlays (Deep Frame-Level Insight) MorphAI v1.5 introduces three specialized overlays that collectively transform the engine into a visual analytics platform for facial computation. The Face Mesh overlay renders a high-density polygonal topology mapped directly onto each detected face, enabling ultra-precise tracking of subtle surface movements and ensuring anatomically consistent deformation throughout morph transitions. The Landmarks overlay extracts and visualizes 68 critical facial anchor points, providing the structural backbone for expression analysis, feature alignment, and stable morphing across diverse face geometries. The Heatmap overlay renders temporal velocity gradients across facial regions, highlighting areas of motion, quantifying the speed of surface shifts, and revealing micro-expressions that would otherwise remain invisible, elevating MorphAI into a true motion intelligence system. 5. Export and Keyboard Shortcuts (Workflow Integration) Morphed outputs can be captured and downloaded directly from the interface, extending the system's utility beyond the live rendering environment into broader production and research workflows. An integrated keyboard shortcut system activates upon opening the interface, enabling rapid control and streamlined interaction for power users operating under time-sensitive or high-throughput conditions. Usage Recommendations and Privacy Considerations For optimal morphing, input images should contain faces that are clearly visible, well-lit, and centrally positioned within the frame. These conditions maximize landmark detection accuracy, bounding box stability, and overall output quality. All images processed through the interface are handled entirely in-session and are never stored or persisted beyond the active session, ensuring strong privacy guarantees for all users. Safety Considerations MorphAI v1.5 retains integrated NSFW content detection, automatically identifying and restricting explicit or inappropriate imagery from entering the processing pipeline. Blocked content is clearly flagged through on-screen notifications, maintaining the system's suitability for academic, professional, and creative contexts.
Stelvin Saji Stelvin (Tue,) studied this question.