Early prototype for magic mirror app -- ideas wanted!

[Sully]

I’m putting together a silly “magic mirror” for a halloween party… I’ve got a bunch of ideas on the actual hardware of the device, still sorting that all out and I’m hopeful I’ll end up with some kind of working device with camera and speakers that at least somewhat resembles a magic mirror prop.

I put together a quick prototype app for the basic functionality: https://thevoid.streamlit.app/mirror

So far you can take a picture, and it starts a chat session with the mirror by asking “Mirror mirror on the wall, who’s the fairest of them all”. It uses Claude 3 Haiku to get a description of the picture, since OpenAI usually refuses to do anything with pictures of people, then the chat is with GPT4 Omni Mini.

The mirror responses are spoken back-- in the actual party version I’ll be using a really nice voice I found on Elevenlabs that sounds a bit like Hermione Granger from the Harry Potter movies, but for this online version I’m just using the python package gTTS so it doesn’t sound very good, but honestly not terrible.

Anyone have any feature ideas? What else should I do / how can I make it a fun Party thing? Open to all suggestions good and bad at this point

You can hover over the Reset button to see the description Haiku gave of the photo, which is usually fun and amusing to me:

edit: Just want to add that none of the images are saved (by me at least), but are sent to AWS Bedrock to be processed by Claude 3 Haiku.

[Jessica_Smith]

This is such a fun idea—how did it go over at the party?

[Sully]

I was not able to get a hardware device built like I wanted, though I’m still working on it. I ended up using a tablet with a customized version of the Streamlit app, running full screen and it was really quite nice. The party version features a voice from 11 labs that sounded really good and I was using the new audio input along with Whisper instead of text input and that worked out really well.

Thanks for checking it out!