In Action
Platform Interface
A glimpse into the real-time AR nail visualization experience — from hand detection to design overlay.
Case Study
A computer vision–driven application that enables users to virtually apply and visualize nail designs in real time using their device camera — powered by a unified YOLOv8 multi-task model.
10
Nails Tracked
Real-Time
Rendering Speed
Multi-Task
Model Architecture
Mobile
Edge Optimized
In the beauty industry, customers cannot accurately visualize nail designs before application — creating friction, lost revenue, and poor experiences across every touchpoint.
Core Problem
Physical trials are costly, time-consuming, and limited in variety — holding the entire industry back.
"Slow decisions drain chair time"
Physical nail trials require significant chair time for both technicians and customers, slowing down service throughput and limiting bookings per day.
"Rework compounds lost revenue"
When customers change their minds after application begins, salons incur costly rework, wasted materials, and lost revenue that compounds over time.
"Static swatches fail customers"
Customers struggle to visualize designs from static catalogs or swatches, limiting confidence in exploring new styles beyond familiar choices.
"Misaligned expectations hurt satisfaction"
Nail salons face persistent inefficiencies in translating customer visions into applied designs, leading to mismatched expectations and dissatisfaction.
"Uncertainty kills online conversions"
Online shoppers lack confidence when purchasing nail products due to uncertainty about how colors or styles will appear on their actual hands.
"Poor tools cost revenue industry-wide"
Poor visualization tools lead to abandoned purchases, increased product returns, and significantly lower conversion rates across in-salon and e-commerce channels.
iOPTIME proposes an AI-powered Augmented Reality system capable of real-time nail detection, segmentation, and tracking — optimized for mobile and edge devices. The core is a unified deep learning model that performs multiple vision tasks simultaneously, reducing computational overhead and improving real-time performance.
A single customized YOLOv8 architecture simultaneously performs nail segmentation and keypoint tracking, eliminating the overhead of separate models.
Deep learning-based nail segmentation produces precise pixel-level masks, ensuring nail designs are applied with sub-pixel accuracy and realistic texture mapping.
Designed from the ground up for resource-constrained devices, the model minimizes inference latency and memory consumption for smooth real-time performance.
Real-time AR nail visualization powered by YOLOv8 SegPose
Comprehensive computer vision capabilities delivering an accurate, interactive, and immersive AR nail experience.
10 nails
tracked simultaneously
Detects and segments all nails in real time with high accuracy, producing precise pixel-level masks across various hand poses and lighting conditions.
Sub-px
tracking precision
Tracks stable spatial reference points on each nail across frames, maintaining alignment and preventing design drift as the hand moves.
0ms
perceived switching lag
Users switch colors, textures, and decorative elements in real time — zero lag, zero reloading.
100%
contour accuracy
Virtual nail designs are applied with sub-pixel precision, ensuring realistic texture mapping that faithfully follows nail contours and curvature.
Real-time
frame-level stability
Smooth, lag-free design visualization maintained frame-to-frame as the hand moves naturally in any direction.
Edge
no cloud required
Reduced inference latency and low memory footprint enable full deployment on smartphones and edge devices without a cloud dependency.
A unified multi-task pipeline from camera input to AR output — built for speed, accuracy, and real-time performance on mobile devices.
Live Feed Capture
Device camera streams live footage at native resolution. Frames are normalized and a hand region-of-interest is cropped before passing to the model — minimizing wasted compute.
Unified Multi-Task Model
A customized YOLOv8 architecture with a dual output head — simultaneously running nail segmentation and keypoint estimation in a single forward pass, eliminating model redundancy.
Segmentation + Keypoints
Two parallel branches decode the shared backbone features independently — a segmentation head producing pixel-level nail masks and a keypoint head generating stable spatial coordinates.
Real-Time Design Overlay
Applies the selected virtual nail design using the segmentation mask for boundary-accurate placement and keypoints for sub-pixel alignment — rendered at display frame rate.
Final Output
Real-Time AR Nail Visualization
Rendered at display frame rate on device
Measurable improvements in salon efficiency, customer satisfaction, and sales conversion through accurate, real-time AR visualization.
Customers compare multiple nail designs interactively in real time, dramatically reducing the time spent on design selection during consultations.
Accurate, real-time visualization bridges the gap between physical trials and digital experiences, setting a new standard in beauty tech UX.
Salons serve more customers per day with streamlined consultations, less rework, and significantly improved throughput per technician.
Online shoppers gain confidence in purchasing nail products by seeing accurate color and style previews on their own hands before buying.
Complex technical and operational hurdles elegantly solved through innovative computer vision engineering.
Maintaining stable nail design overlay while the hand moves naturally required a robust keypoint tracking system tightly coupled to segmentation output.
Running a multi-task deep learning model at real-time speeds on mobile devices demanded architectural optimizations including quantization and head pruning.
Training data diversity was critical to ensuring accurate nail segmentation across a wide range of skin tones, lighting conditions, and nail lengths.
Ensuring both the segmentation and keypoint heads stay synchronized within the same forward pass required careful loss balancing during multi-task training.
Built from the ground up using modern, production-ready technologies designed for real-time computer vision and edge deployment.
YOLOv8
AI Model
Python
Backend
OpenCV
Vision
PyTorch
Deep Learning
React
Frontend
TypeScript
Frontend
Transformers
AI
ONNX Runtime
Inference
WebRTC
Streaming
FastAPI
API
TensorRT
Optimization
NumPy
Compute
A glimpse into the real-time AR nail visualization experience — from hand detection to design overlay.