Nov 08, 2021 20:00:00

'Hypnogram' that draws an illustration based on the keyword entered by AI

Advances in technology have made it possible for algorithms learned through deep learning to draw pictures. 'Hypnogram' is a site that automatically draws illustrations based on the keywords you enter using AI, and can be used from your browser. I tried entering various keywords to see what kind of pictures it would actually draw.

Hypnogram

https://hypnogram.xyz/

To create an image, just enter a keyword and click 'Generate'. However, the process takes time, so you will need to wait approximately 2 to 3 minutes. This time, we asked AI to create an image using 'GIGAZINE'.

When the illustration is completed, the illustration will be displayed below the input field, and by clicking 'Download' you can save the illustration in 480 x 480 pixel JPEG format. Looking at the illustration, there are two things that look like the 'GIGAZINE' logo, and for some reason it is interesting that there is an illustration of only the lower half of the man's face.

An illustration is created each time you click 'Generate', so a different illustration will be created each time. Below is another illustration created by AI from 'GIGAZINE'. The colors are brighter than the previous illustration, and the string in the center of the screen is closer to the GIGAZINE logo. And it is a mystery that there is a man's face in the center of the image.

This time, I created an illustration by entering the name of a huge tech company, ``Google Amazon Meta Apple''.

The result looks like this. The Google and Apple logos seem to be output fairly well. The 'A' on the left may be somewhat similar to Amazon's A. Perhaps because it has been a while since Meta changed its company name, related elements were not drawn in the image.

According to the tips on the site, if it is not a short keyword but a long phrase that combines multiple words, it is better to change it to 'VQGAN + CLIP' instead of 'Hypnogram'. VQGAN is an image generation model that develops '

VQ-VAE ', which is a VAE using vector quantization (VQ), into a generative adversarial network (GAN), and a neural network 'CLIP' that connects text and images. You can understand the combined efforts by reading the article below.

Will image generation AI change the nature of art? -GIGAZINE

So, I searched for ' Lucy in the Sky with Diamonds ' using VQGAN+CLIP.

The results are below. A lot of diamond-like jewels are displayed in the starry sky, and the severed head of a man who looks like a member of

the Beatles who sang the song 'Lucy in the Sky with Diamonds' is floating on a cloud.

In addition, as a creation example, the work and keyword were displayed as a set at the bottom of the page, so I looked at various things.

'Heaven on Earth'

'something'

'Girl eating ice cream'

'tits'

There was another work depicting 'breasts', but in both of them the contours of what appeared to be the lower half of a woman's face and round skin-colored objects were emphasized, making it hard to call them breasts. It seems like it will take a little more time for AI to understand breasts.