How well does GLIDE: the text-to-image generator work?

On December 22, 2022, Open AI released GLIDE, an intriguing artificial intelligence. If you give a short sentence, GLIDE will generate an image for you. I found an excellent article below.

I ran GLIDE with colab, a convenient tiny Google development environment. And I will introduce the results on this page. GLIDE generated all images on this page from the short sentences associated with each picture.

First of all, GLIDE is not familiar with Albert Einstein nor Steve Jobs, but GLIDE speculated that Albert or Steve might be the name to give to cats or dogs. See (1a, 1b.) I can’t blame GLIDE for this. However, GLIDE looks to believe Steve Jobs is cleverer than Albert. See (1c.)

I listened to more of GLIDE’s private life. He thinks he is a wise dog-like animal (2a.) And he also showed me a picture of his lover (2b.) The one who created him seems to be a great calculator rather than a giga-scaled computing machine or human being (2c.) He is humble than I expected.

GLIDE knows human beings eat hamburgers. Especially a hamburger shop in San Mateo looks like his favorite (3a, 3b, 3c.)

GLIDE has a sense of the season. Not only that, but GLIDE also understands the differences between narrow trails and winding roads (4a, 4b, 4c.)

Conclusion

The GLIDE showed the practicality of the “diffusion model.” At the same time, many application fields will open for the group of picture-creating artificial intelligence. Have fun.

Tokyo-life hacker / doing yoga in daily routine