Custom YOLO models trained on specific games outperform general models significantly. Here's how to train your own.
Step 1: Collect Screenshots
Play your target game and capture 200-500 screenshots with enemies visible. Use different maps, lighting conditions, and character skins. The more variety, the better.
Step 2: Label Your Data
Use a tool like LabelImg or Roboflow to draw bounding boxes around enemy players in each screenshot. Label them as "player" or "enemy." This creates the training annotations.
Step 3: Train the Model
Use Ultralytics YOLOv8 or v11. Training on 300 images takes about 30-60 minutes on a modern GPU. Use Nano size for maximum inference speed.
Step 4: Export to ONNX
Export your trained model to .onnx format. Skyrant loads any .onnx file directly — it auto-detects resolution, classes, format, and precision.
Step 5: Load in Skyrant
Upload your .onnx file through the Skyrant dashboard. The client downloads it automatically and starts using it for detection. No configuration needed.