Google’s 20 Billion Parameter AI Image Generator Is Insane

Jim Clyde Monge
CodeX
Published in
4 min readJul 5, 2022

--

Screenshot from Google’s Parti Research

Text-to-image AI generation models are very popular these days.

When OpenAI announced Dall-E2, the internet was in frenzy.

Shortly after, Google announced that it is building its own text-to-image AI tool called Imagen.

But even before Imagen reaches a private beta phase, Google announced another AI tool that basically does the same thing — renders hyperrealistic images from short textual inputs.

The new tool is called Parti which stands for Pathways Autoregressive Text-to-Image.

Let’s talk about a few things.

  • What is Google Parti and how does it work?
  • How is it different from Google Imagen?
  • How does it compare to OpenAI’s Dall-E2 and MidJourney?

What is Google Parti?

Parti is a text-to-image AI generation model that takes in a simple text prompt from a user and generates hyperrealistic images with its encoder-decoder that can scale up to 20…

--

--