Imagine an AI that not only describes images but also fact-checks its own descriptions, ensuring remarkable accuracy and detail. That's the promise of Visual Fact Checker (VFC), a new research project from NVIDIA. Unlike typical image captioning AI, which can sometimes hallucinate details or miss key information, VFC employs a unique three-step process. First, it proposes multiple initial captions. Then, it uses other AI tools like object detection and visual question answering to verify the accuracy of those captions. Finally, it synthesizes the verified information into a polished, high-fidelity description. This approach allows VFC to generate captions that are both detailed and accurate, even for complex 3D objects. The researchers tested VFC against existing captioning models and found it produced superior results, rivaling even proprietary models like GPT-4V. They also introduced a new metric, the CLIP-Image-Score, which measures caption quality by comparing the original image to an image reconstructed from the AI-generated caption. This innovative approach opens exciting possibilities for applications like image search, accessibility for visually impaired users, and content creation. While challenges remain, VFC represents a significant step towards more reliable and detailed AI-generated descriptions of the visual world.