m87-labs/moondream ? reverse-engineered prompt

Build me an application that uses the moondream vision model. I want a simple web UI, maybe using Gradio since I see some demos for it, where I can experiment with its capabilities.

The main feature should be letting me upload an image and then ask the AI questions about it. For example, I should be able to upload a picture of a room and ask "what color is the couch?" or "what is the person on the left doing?". The goal is to have a chat-like experience based on the image I provided.

I also saw a file for a webcam demo. Can you please add a second mode to the app that connects to my computer's webcam and gives me a live, text-based description of what it sees in real-time? The README says this model is small and efficient, so it should run pretty smoothly on my local machine.