AI powered face liveness for apps that need spoof checks.
TinyLiveness is a lightweight passive RGB model that answers one question: does this aligned face crop look like a live person, or a presentation attack such as a photo, video, or screen replay?
The API prediction model is the APCER 1% variant
The default public API route uses `tinyliveness_main_apcer1_224.onnx`. This is the strictest release policy and is meant to minimize spoof acceptance, even when that creates more legitimate-user friction.
- Model
- EfficientNet-B0
- Input
- 224 RGB
- Runtime
- ONNX API
- Policy
- APCER 1%
MIT licensed so anyone can use it
TinyLiveness is released under the MIT License. You can use it, modify it, copy it, distribute it, and build commercial or non-commercial applications with it, as long as you keep the MIT copyright and license notice.
- MIT License
- Python backend route included
- JavaScript website tester included
- FP32 ONNX model artifacts included
What TinyLiveness is, and what it is not
This is a passive single-frame RGB liveness signal. It should be combined with face detection, alignment, identity matching, rate limits, logging, manual review, and product-specific fraud controls.
| Use case | Good fit? | Notes |
|---|---|---|
| Open-source liveness API | Yes | Open source MIT license completely free to use |
| Kiosk or onboarding pre-check | Yes, with controls | Run the model before face matching and add retry/manual-review paths. |
| Standalone fraud decision | Depends | Liveness should be one security signal, not the full decision engine. |
| Certified PAD compliance | No | Formal PAD claims require lab testing and broader datasets which we are working on. |