OpenGVLab/InternVL ? reverse-engineered prompt

Build me a working open source multimodal AI project like InternVL, where I can chat with images and get useful answers, and also try a few vision tasks like image classification, segmentation, and video understanding. I want a clean local demo I can run from the repo, with a simple web interface for uploading an image or video, asking questions, and seeing the model response. Please include the training and inference code structure in a way that feels complete, so it’s easy to swap in different model sizes later. If you need to check current setup details or best practices, look up the latest docs online. Make it feel polished, with clear install steps, a basic example, and sensible defaults so it works without a lot of manual setup.