Exploring the World of AI Art Generators

Case Study: Anjali

For this case study, I created a self portrait using the AI generator called Dalle-2. I typed in prompts describing myself in order to try to create an image close enough to the way I look.
I started off with a very basic outline of how I wanted my image to look.

1. Prompt: Portrait of girl with black hair, brown eyes, indian anime

I wanted to add in an art style as initially, my first prompts did not have a style and bits and pieces of the images that came out seemed to be copied from different images on the internet.

2. Prompt:Portrait of girl with black long hair and brown eyes, digital art style

I tried getting into a little bit more detail in my next prompt. I made the style digital art. However this seemed a bit general. I also found that in the digital art style, if not specified, the eyes tend to be large.

3. Prompt: Portrait of girl with black long hair and brown small eyes, indian ethnicity, digital art. An illustration by Leonardo da Vinci

In the next prompt, I experimented a little by adding an artist's style (Leonardo da Vinci). I tried adding in the ethnicity as well. However the image that came out was a very traditional one.

4. Prompt:Portrait of girl with black long hair and brown small eyes, digital art. An illustration by Leonardo da Vinci

By removing only the ethnicity and generating the image, I got a completely different outcome.

5. Prompt: An illustration by Leonardo da Vinci portrait of a woman in her early 20’s with black long hair and brown eyes, thick eyebrows, slightly upturned nose,digital art with a textured background from the indian subcontinent.

The next prompt was slightly more detailed.

6. Prompt: An illustration by Leonardo da Vinci portrait of a woman in her early 20’s with black long wavy hair, brown small eyes with eyeliner, thick eyebrows, square jaw and a slightly upturned nose ,digital art with a textured background from the indian subcontinent.

Finally, I added in more detailing in the face structure and the hair and nose in order to reach an image that looks slightly similar to me.

Conclusions reached by using Dalle-2:
1. The more detailed the prompt is, the more accurate the image will be.
2. Everytime I added a new slight detail into the prompt, the image that was generated became completely different. I was hoping that the overall image would be retained and only the details would be changed; however, this did not occur.

-Anjali Rao