Gigglygram
About the project
You're not funny, but AI is! GigglyGram uses generative AI to turn you into a witty conversationalist with clever comebacks and memes.
Project info
Difficulty: Moderate
Platforms: NVIDIA
Estimated time: 7 hours
License: GNU General Public License, version 3 or later (GPL3+)
Items used in this project
Story
You're not funny, but AI is! GigglyGram is a messaging app that uses generative AI to turn you into a witty conversationalist with clever comebacks and amusing memes.
How It WorksI built a quick-and-dirty proof of concept Android messaging app with the help of the MIT App Inventor. This app can receive text messages, and take some actions based upon them.
There is a "Make Me Funny" button that feeds the text of the most recently received message into an open source Large Language Model (LLM), along with some instructions telling it to provide a funny response to the message. The LLM is packaged up as a llamafile, and I specifically used the seven billion parameter Mistral-7B-Instruct model. This runs on an NVIDIA Jetson AGX Xavier Development Kit, and is accessed by a simple web API written in Python.
There is also a "Meme" button that leverages the same web API to send a prompt to a Stable Diffusion 2.1 image generation model running on the Jetson. The prompt asks the model to generate a funny meme based on the latest text message received.
MediaA message was received:
A suggested response was generated by the LLM:
The meme was generated by Stable Diffusion 2.1:
An example meme generated during testing:
Bill of Materials- 1 x NVIDIA Jetson AGX Xavier Development Kit
Leave your feedback...