A blog related to original AI artworks. I make AI artworks using coined words and emoji as prompts.

I tried to draw AI art by playing with the spelling of the artist’s name

I gave only “Picasso”, “Dali”, and slightly different spellings on the AI ​​art site, and tried to see how the generated images would change. Sometimes an image that is completely different from the paintings of those painters is created, but sometimes a painting-like image is created that is different from the spelling. As for AI art, it seems to be common to give a long text to draw the picture you want, but I think it is also interesting to give a snippet of text like this.

If you give Stable Diffusion 1.5 the words “Picasso” or “Dali”, it will usually output a pictorial image that looks like it was drawn by those artists. For example, for “Picasso”, the picture looks like this.

With “Dali”, for example, you can get the following picture.

So, first, let’s change the spelling of “Picasso” a little. “Pocasso” gives the following picture.

I think that it doesn’t look like Picasso so much, but I’m trying to paint like a modern painter at least. Depending on the spelling, it may not look like a painting. For example, “Picassi” generates only pictures that look like soccer pictures. Is it because “Picassi” is in the name of a football player (I don’t know)?

Select two images from “Pucasso”, “Picasso”, and “Pcasso” and put them on.

Although they are separated from Picasso, they are painted in the style of modern painters.

Next, let’s change “Dali” a little. First, let’s put “Dalí”, the original spelling of Dali. This seems to tend to be interpreted as Dali himself rather than a painting of Dali. However, the reason why such images are created is probably that I see many photographs of Dalí’s own performances during my studies.

Of these, only one is in the style of a Dali painting.

Spellings such as “Dal”, “Dalo”, and “Dalli” don’t look like paintings, so put two images each of which is drawn with “Dalì”, “Dalî”, and “Dalï”.

Even with these spellings, I sometimes get a photographic image of “Dalí”, but I will omit it.

Here, I tried “Dalï” using DALL·E 2 instead of Stable Diffusion. I got completely different results. I had them make four pictures, and they all looked like photographs of Dali’s dolls. Here are two of them.

To reiterate, it seems common for AI art to give long texts to make it draw the picture you want, but I think it is interesting to give just-scraped texts like this.

Note: This article is a revised and translated version of an article in Brog from Kanada (in Japanese).