Listens for a command in a Twitch Channels chat, sends to OpenAI API, generate vocalized response.
Originally a fork from https://github.com/Pedrolian/twitch-obs-scene-changer But has become its own thing.
This will listen to a twitch channel's chat for the !talk command. It will take everything after that and send it to OpenAI's API. The response will be saved in a text file, then converted to speech and automatically played back.
I has some pretty strict twitch restrictions and people still found ways to circumvent the filters and have the AI say pretty despicable things. It did result in a twitch channel ban. I would recommend adding some error checking and input sanitization into the program itself.
- NodeJS - NodeJS v.10
- OBS-Websocket Plugin
- OpenAI API Key
- Python
Run npm install to download dependencies.
npm install
Install the OpenAI package
pip install openai
Install VLC player
https://www.videolan.org/vlc/
Configure OBS.
Tools -> Websocket server settings
Check "Enable Websocket server"
Check "Enable authentication"
Enter a password
Configure .env file
TWITCH_CHANNEL="" // <-- Your twitch username
OPENAI_API_KEY="KEY" // <-- OpenAI API Key
OBS_URL="localhost:4444" // <-- OBS Plugin url. "localhost:4444" is the default url:port if connecting to OBS locally and using default port in plugin
OBS_PASSWORD="" // <-- OBS Plugin password
Configure line 7 on dynamic.js if you want to use the "!talk " command to a different command name
const COMMAND_NAME = "!talk"; // Command to type in chat.
Configure OpenAI API key in text_to_speech.py
openai.api_key = "KEY"
To run dynamic.js in cmd promt / terminal
node dynamic.js
I created the background images, logos, and presenter images with Midjourney https://www.midjourney.com/
To animate them, I used D-ID https://www.d-id.com/
Create an animated face at rest, and a second video of text to speech. Alternatively you could create a paid account and connect requests to the D-ID API.
Create a scene in OBS at rest and a Scene with the speaking face.
Configure OBS to change scenes when sound is active.
