How Far Can VLMs Go for Visual Bug Detection? Studying 19,738 Keyframes from 41 Hours of Gameplay Videos

Video-based quality assurance (QA) for long-form gameplay video is labor-intensive and error-prone, yet valuable for assessing game stability and visual correctness over extended play sessions. Vision language models (VLMs) promise general-purpose visual reasoning capabilities and thus appear attractive for detecting visual bugs directly from video frames. Recent benchmarks suggest that VLMs can achieve promising results in detecting visual glitches on curated datasets. Building on these findings, we conduct a real-world study using industrial QA gameplay videos to evaluate how well VLMs perform in practical scenarios. Our study samples keyframes from long gameplay videos and asks a VLM whether each keyframe contains a bug. Starting from a single-prompt baseline, the model achieves a precision of 0.50 and an accuracy of 0.72. We then examine two common enhancement strategies used to improve VLM performance without fine-tuning: (1) a secondary judge model that re-evaluates VLM outputs, and (2) metadata-augmented prompting through the retrieval of prior bug reports. Across \textbf{100 videos} totaling \textbf{41 hours} and \textbf{19,738 keyframes}, these strategies provide only marginal improvements over the simple baseline, while introducing additional computational cost and output variance. Our findings indicate that off-the-shelf VLMs are already capable of detecting a certain range of visual bugs in QA gameplay videos, but further progress likely requires hybrid approaches that better separate textual and visual anomaly detection.

SUPERNOVA: Automating Test Selection and Defect Prevention in AAA Video Games Using Risk Based Testing and Machine Learning

Testing video games is an increasingly difficult task as traditional methods fail to scale with growing software systems. Manual testing is a very labor-intensive process, and therefore quickly becomes cost prohibitive. Using scripts for automated testing is affordable, however scripts are ineffective in non-deterministic environments, and knowing when to run each test is another problem altogether. The modern game's complexity, scope, and player expectations are rapidly increasing where quality control is a big portion of the production cost and delivery risk. Reducing this risk and making production happen is a big challenge for the industry currently. To keep production costs realistic up-to and after release, we are focusing on preventive quality assurance tactics alongside testing and data analysis automation. We present SUPERNOVA (Selection of tests and Universal defect Prevention in External Repositories for Novel Objective Verification of software Anomalies), a system responsible for test selection and defect prevention while also functioning as an automation hub. By integrating data analysis functionality with machine and deep learning capability, SUPERNOVA assists quality assurance testers in finding bugs and developers in reducing defects, which improves stability during the production cycle and keeps testing costs under control. The direct impact of this has been observed to be a reduction in 55% or more testing hours for an undisclosed sports game title that has shipped, which was using these test selection optimizations. Furthermore, using risk scores generated by a semi-supervised machine learning model, we are able to detect with 71% precision and 77% recall the probability of a change-list being bug inducing, and provide a detailed breakdown of this inference to developers. These efforts improve workflow and reduce testing hours required on game titles in development.