This project is an ambitious one, where we want to use some of the most questionable parts of the internet, comment sections and chatrooms, to create art.

We also want to make the project easily accessible so anyone can try their hand at generating some fine AI art. If you can't wait to try it out yourself, here's a link to the project.

Supplies

Raspberry Pi
DeepAI
Remo.tv

Step 1: Project Video

Step 2: Remo.tv

The first step is gathering chat messages and comments. With our idea to make it easily accessible, Remo.tv is a natural pick. It's a robot streaming platform which allows you to connect all kinds of hardware to the internet, and let anyone control them. It also has a chat functionality and the ability to display images, which is exactly what we're looking for!

In this, case the hardware we'll be using is a Raspberry Pi.

Remo.tv has a great Github page with setup instructions.

Once setup, our Raspberry Pi can start receiving the chat messages, sent via Remo.tv.

Step 3: DeepAI

With Remo.tv setup we can move on the artsy part. Each comment that we receive needs to be transformed into art, and to achieve this we'll be using some artificial intelligence magic.

Luckily there as another platform to make our life easy, DeepAI. They have all kinds of AI related features, but the ones we are interested in are their APIs.

The first API we use is Text To Image, all we need to do is send a text and wait for the magic to happen. In the above picture you can see the result of sending A dog with a funny hat.

Our generated picture isn't art just yet, so we use their Fast Style Transfer. This API expects an original image, in our case our generated one, and a style to apply. You can see the result of combining our dog with a funny hat and a classic van Gogh painting.

Step 4: Dataflow & Code

With al the separated pieces completed we can connect them. In the drawing we give an overview of the data flow:

A chat message arrives from Remo.tv to our Raspberry Pi
Our Pi sends this message to the Text To Image API and receives a generated image back
This image, together with a randomly selected art style, is then sent to the Fast Style Transfer API
After receiving the combination of the art style and the generated image, the Raspberry Pi streams the result to Remo.tv.

To be able to stream the generated image to Remo.tv we needed to write some custom code. Luckily, the lovely Remo.tv community helped us with that, thanks guys! :)

For all the curious, the full code is included in this project so you can get started right away.