AI TV

The idea was to create an online livestream that generates copyright free TV content using AI. The livestream would also generate content based on user comments.

I created a prototype and tested it on Twitch, but some of the content is inappropriate and not presentable so I won't show it here.

How it works:

The livestream is basically a 3D Unity project managing network requests between GPT 3.5, Amazon Polly and Twitch API.

1. A prompt is sent to GPT 3.5. The prompt is selected using characters and topics recommended from Twitch chat. Overall, it will be something like "Generate a TV script containing dialogue between characters: Bob and Frank. The topic is: the weather".

2. After GPT 3.5 responds the incoming script is parsed. Each line of dialogue for each character is sent to Amazon Polly to generate the corresponding audio.

3. After Amazon Polly responds with the audio, the 3D character models are retrieved and placed in the scene. The camera changes focus to whoever is speaking according to the script and the newly retrieved audio is played, thus simulating a back and forth exchange between 2 characters like in a TV show.