A simple API call returns an age estimate and gender prediction in under 500ms. Built for age verification, visitor analytics, and identity insight.
Send a face image, get back structured JSON — no setup beyond an API key.
Continuous age with MAE 3.555 on LAGENDA 84k — beats MiVOLO v2's face+body baseline using face-only input.
97.75% gender accuracy from a single face crop — no body required, no extra configuration.
Results in under 500ms. Optimised inference pipeline for production throughput on GPU and CPU.
All inference endpoints are API-key protected over HTTPS. Keys are issued per integration.
Works with any language that can make an HTTP POST. No SDKs required.
Wherever age and identity verification matters, TellMyAge fits in.
Enforce age gates for gambling, alcohol, and adult content platforms in real time.
Integrate with ClientScan's self-exclusion portal to flag excluded individuals at entry.
Understand your audience's demographic profile from CCTV or in-store cameras.
Cross-reference estimated age against declared DOB for soft identity verification workflows.
Answers to the most common questions about the API, accuracy, and integration.
On the LAGENDA 84k public benchmark, FaceAge ClientScan achieves a mean absolute error (MAE) of 3.555 years, with 75.5% of predictions within 5 years of the true age (CS@5) and 97.75% gender accuracy — using only the face crop, outperforming MiVOLO v2 which requires both face and body bounding boxes.
Yes. TellMyAge is built for age-restricted commerce, gambling platforms, alcohol delivery, adult content gates, and self-exclusion enforcement in regulated industries.
Median inference completes in under 500 milliseconds per request. GPU-backed deployments serve hundreds of concurrent requests per second with sub-100ms per-request latency.
Yes. A free interactive demo is hosted on HuggingFace Spaces where you can upload a face photo and receive age and gender predictions instantly.
TellMyAge is powered by FaceAge ClientScan, a DINOv3-ViT-L/16 backbone (307M parameters, pretrained on LVD-1.68B) fine-tuned in 4 phases on 4M face images. Age uses CORAL ordinal regression (age = Σ sigmoid over 100 logits); gender uses a 2-way softmax head. Production model is exported to ONNX for optimised inference on CPU and GPU.
FaceAge ClientScan achieves MAE 3.555 on LAGENDA 84k using only the face crop, beating MiVOLO v2's paper claim of MAE 3.650 which requires both face and body bounding boxes. It wins 6 of 7 age groups against MiVOLO v2, with the largest improvement in the 66+ group (−1.297 years).
Yes. Evaluated across 8 external benchmarks (UTKFace, IMDB, MORPH, AFAD, CACD, FG-NET, APPA, AgeDB), the average MAE is 4.884 years, with the best score of 3.520 on AFAD and 4.235 on MORPH.
Input is a 224×224 RGB face crop with ImageNet normalization (mean 0.485/0.456/0.406, std 0.229/0.224/0.225). The face bounding box must be padded by 10% proportionally on each side before cropping — this matches the training setup and is required to reproduce MAE 3.555.
The FaceAge ClientScan model weights are released under the Apache 2.0 license on HuggingFace. Commercial use is permitted with attribution to Trung Thanh Tran and ClientScan Limited.
Yes. TellMyAge serves requests globally over HTTPS. The service is operated by ClientScan Limited (United Kingdom) and complies with GDPR for European traffic.
Reach out about integrations, pricing, or enterprise support.
7 Bell Yard, London, WC2A 2JR, United Kingdom
ClientScan Limited — clientscan.co.uk
UK00004093000