top of page

Baby daikon radish in a tutu walking a dog!

Photo Courtesy:

Believe it or not, that is a Machine's point of view on how a baby daikon radish in a tutu walking a dog might look like.

If I have to be honest, I don't think I could have done a better job than that - granted I am awful at even stick figures!

What you are seeing above is the output of the Dall-E ML Algorithm that can generate images from input text. The first version of this algorithm came out in early 2021. Recently, has released a newer version of the algorithm called Dall-E 2 which can make realistic edits to existing images from a natural language caption. My favorite output was this one..

When I entered the text: Teddy Bears, shopping for groceries, in Ancient Egypt!!

The algorithm generated this image. It's very photo realistic.!

Photo Courtesy:

The model is trained on large number of images and their text descriptions. Not only does it understand objects (bears and grocery shopping) but also the relationships between objects (bears grocery shopping).

In simple terms, the training dataset consists of pairs of images and their corresponding captions.

There are 2 steps to the process:

  1. An encoder which take a prior image, and generates its CLIP image and text embeddings.

  2. A decoder that produces images conditioned on CLIP image embeddings and optionally text captions.

The results show that explicitly generating image representations improves image diversity with minimal loss in photorealism and caption similarity.

These models are, of course, just the beginning of a really exciting chapter in Machine Learning. Try it out for yourself here and tell us your thoughts in the comments section!

19 views0 comments

Recent Posts

See All


bottom of page