ClearShot

Clear sharp images are the key to great datasets, and great AI models. ClearShot is a powerful, hardware-accelerated desktop application designed to streamline the creation of AI/ML training datasets. By combining a blazing-fast Python backend with a sleek, minimalist React interface, it allows researchers and creators to effortlessly extract high-quality, perfectly squared face and body crops from both local videos and YouTube links. Under the hood, ClearShot leverages ONNX Runtime—with full Metal GPU support for Apple Silicon—to run state-of-the-art detection models (like SCRFD and YOLOv8) at lightning speeds. With intelligent built-in features like Variance of Laplacian blur-rejection and perceptual hash deduplication, ClearShot guarantees that only the crispest, most unique frames make it into your final training data, saving you hours of manual curation. This project runs effectively on an M1 MacBook Air and all Apple chipset generations released after.

Working on CUDA support for Nvidia GPUs for Linux and Windows. Also working on a native Swift version for simpler deployment.

Checkout and try the project by downloading it from Github.
The project is created with an MIT license so feel free to use it to your heart’s content.