NiceShot AI: Computer vision tool auto-detects gameplay events in Call of Duty

NiceShot AI is a Python-based analytics tool designed to automatically analyze Call of Duty gameplay videos using computer vision. The tool is powered by YOLOv8n (fine-tuned on custom collected and annotated gaming video datasets under CC license), OpenCV, YOLO, FFmpeg, and matplotlib, enabling it to detect, track, and extract key in-game events.

The tool supports three Call of Duty titles: Call of Duty: Black Ops 6 (2024), Call of Duty: Modern Warfare II (2022), and Call of Duty: Black Ops 7 (2025, still in testing). For each game, it detects specific events. In Black Ops 6, it identifies gun kills (face-to-face only), medals (when earned by the player), and deaths, though medal type is not differentiated—only the count is recorded. Modern Warfare II support is also in testing, with kill detection and death detection available without the face-to-face limitation. Black Ops 7 detection is likewise in testing with kills and medals tracked.

Key features include event timestamping and CSV export with two columns (Timestamp and Event) for further analysis, automatic clipping of detected events, and session analysis reports featuring multiple charts for post-session stats review. The tool uses EasyOCR to prevent false event counts during special game scenes like killcams and spectating modes. Special events such as kill streaks—detected from the combination of multiple consecutive kills within a time threshold—are also tracked. Auto-clipping exports in both 16:9 and TikTok vertical formats, and the tool can concatenate clips into highlight reels of custom lengths with fade transitions. Bulk analysis of Twitch streams is in testing, and clips can be ranked by special criteria (for example, "hot kill clips" where multiple medals appear during a single event).

Processing speed varies by hardware. On a laptop with RTX5070 8GB, i9 14th gen CPU, and 32GB RAM, the tool reaches up to 170 FPS with frames_to_skip = 5. On a laptop with GTX1650 4GB, i7 10th gen, and 16GB RAM, it achieves up to 60 FPS under the same settings. Advanced OCR-based detection temporarily reduces speed from 170 FPS to 30 FPS while confirming events; this feature can be disabled to preserve speed at the cost of potential false positives during spectating scenes. The creator acknowledges that event detection is not perfect—events can be detected more than once or not at all. Installation requires FFmpeg, PyTorch (with CUDA support, the creator used CUDA 12.1 for GTX1650 and nightly CUDA 12.8 for RTX5070), and Python 3.10.11 or compatible. A demo video is available showcasing the tool's results.