Try at home: Generating a profile picture with DALL-E

Benoit Lamouche
5 min readJan 17, 2024
This is the AI version of me

I embarked on a personal experiment with DALL-E. The question is simple: Is it possible to generate a profile picture that resembles reality?

First of all, here is what I look like for those who don’t know me:

Test 1 : Find a prompt that yields a satisfying result

Spoiler: It took me many attempts to arrive at something unsatisfactory.

Here is the very first prompt I gave:

1:1 photography, profile picture of myself.
I’m 1m94 tall, my weight is 80kg, I’m french, my skin type is european, I have 5mm short brown hairs, I have a 3 days beard.
I want to me be smiling a little bit, dressed with a black tee-shirt, in a warm environment.

And the result :

Here is the last prompt I gave:

1:1 photography
profile picture of a guy
1m94 tall
40 years old
weight is 80kg
french, skin type is european, almost no tan
very short hair cut with clippers on full head, dark hairs
1 day beard
little bit tired face
smiling a little bit
dressed with a black tee-shirt
athletic and lean body
facing the camera

And the result :

Between the two, I made about ten iterations trying to refine the description without success. It’s the hair where it struggles the most, a shaved head is not in its codes.

I am honestly disappointed with the result, I sincerely believed I could achieve something more lifelike.

Test 2 : Generate a description from a photo, then generate an image from this description

My prompt :

give me a precise description of this person

Result :

The individual in the image is a man appearing to be in his 30s. He has short brown hair, light stubble on his face, and brown eyes. He is wearing a blue t-shirt and has a neutral expression on his face. The background is plain white, and he is looking directly at the camera. His posture is upright, suggesting a front-facing portrait-style photograph.
The man in the image has a medium complexion with an oval face and a straight, medium-length nose. His short brown hair is neatly cropped and appears to be thick. The man's eyebrows are of average thickness and match the color of his hair. He has brown eyes that are almond-shaped and set evenly on his face, with a direct and level gaze. There is a light stubble on his chin, above the lip, and along his jawline, suggesting a day or two of beard growth, providing a rugged texture to his otherwise clean appearance.
The man's lips are closed and evenly proportioned, and his ears are of average size, aligned with his eyes. The expression on his face is neutral, with no particular emphasis on any feature, and there is a slight crease visible on his forehead. His blue t-shirt has a crew neck and fits comfortably, neither too tight nor too loose.
The lighting in the image is even, with no harsh shadows, indicating a well-lit environment. Overall, his appearance is that of a casual, everyday setting, and the white background provides no distractions, focusing all attention on him.

And the final result when submitting this description :

Again, that’s not great, If I give this picture for a passport, I really doubt about my capacity to travel.

Test 3 : Play with style options

When everything goes wrong, the best thing to do is to play with image styles, an area where DALL-E excels. So, I submitted my profile picture and then asked it to reproduce it in different styles…

In the style of ‘The Simpsons’ animation
In the style of ‘South Park’
In the style of ‘Family Guy’
In a pixel art style
In the style of a 1990s Japanese manga
In the style of a 4-year-old child’s hand drawing

I really doubt the ability of a 4-year-old child to produce a drawing of this quality, if I refer to the benchmark of my 4-year-old daughter…

This is the end of this experiment, whose only objective was to play while exploring a bit more the capabilities of generative AI.

If you liked it, leave some claps, don’t hesitate to test it out for yourself and see if you get more satisfying results. Subscribe below to my other publications:

--

--