This invention identifies AI-generated content in videos by analyzing motion inconsistencies. It works by first identifying objects, people, and backgrounds in each video frame using a CNN model, and then tracking how individual pixels move by dividing frames into patches. If the movement of these elements appears unnatural or inconsistent with their properties, the system flags that part of the video as AI-generated.
Why it matters: Filed as AI video generation was rapidly advancing. The subsequent two years have seen a significant increase in the sophistication and accessibility of AI tools capable of generating highly realistic video, making robust detection methods like this more critical.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI