Tetris, a popular video game with movable shapes, has evolved over time. To maintain its display concept, platforms adapt it to various forms. One of the problems found is how these platforms can consistently maintain the display concept of Tetris. The purpose of the study is to detect and recognize Tetrominoes from Tetris online games using YOLOv9 algorithm models. A total of 93 annotated images were split into three datasets, i.e., training dataset consists of 72 images, validation dataset has 15 images, and testing dataset has 6 images. The result shows that YOLOv9c achieved 84% for the F1-score value and 0.785 mAP on the training data. When the model was tested on the testing dataset, the value dropped to 83%. However, the model correctly counts the number of Tetrominoes in the gameplays. The challenge of detecting a new small and moving image data requires further attention for future improvements.