How Cyclops sees, and how to swap what it sees with. · 4 min read
Cyclops reads the screen with a vision model and never touches a game's memory. That is why it works in anything with a screen, and why its detectability stays low.
The models store lists models trained for particular games and general ones for anything person-shaped. Smaller models run faster; larger ones see further. Each page says what it was trained on, its input size and how many classes it knows.
Press Download on a model and it joins your library. Cyclops's settings page has a Model picker under Detection; pick the one you want there. CUDA and DirectML are used when the machine has them, otherwise the CPU.
Any ONNX detection model works. Put it in the models folder next to the app, or use Add model on the settings page, and it appears in the picker.
