All you need is to provide is some words in a natural language to a bot through what we call a prompt. Then, here we use Midjourney, it generates you something that "feels" your idea. You can even generate realistic photographical images. The more precise you get, better your image is. 

As I'm writing this note, I used the following versions: 

- ChatGPT v3

- Midjourney v5.1

This is already impressive, but they have their limitations and imperfections. We can assume that in the next few months, years, it will drastically improves.

So for today, I have started with the following prompt: 

This is quite detailed already you may say. Midjourney always provides 4 altenates versions, you get to iterate on some of them. When you are happy with you result, you can ask for better resolution one.

It provided me these images​​​​​​​:

The result is quite stunning! Colors and tones give this nice warm feeling. The depth of field is accurate and emphasizes the subject the way we'd expect for this kind of street photo exercice. There is not much glitches to be found. We see the pictures are having some quite realistic render. The pictures are all well structured. You can see some shadows are missing under the bike of the third and. One bike has four cables. To be honest, I have seen worse.

I wanted to see how it behaves when you provide more details.

So I trained ChatGPT. Basically, I showed it prompt examples, explained it how to behave with midjourney's parameters. I asked it to provide me a more detailed scene. 

Here, I'm looking for photorealistic images, so I tought it to also provide some technical camera details. A render from a polaroid differs from a Canon R5. 
I assume also there could a bias from the pictures it was trained with. I.e. Street photographic styles tend not to be shot with polaroids, but more common expensive lens.

if you are interested, I will create a dedicated paper about this

This is the answer ChatGPT filled after my previous prompt. 

Do you find a difference?

I personaly, really find that images convey a more emotional feeling when provided with a more complex prompt. I also really like how the AI made these portrait quite different, they are not so much all the same guys stereotypes. There is variarity.

It seems providing your prompt with deeper info really makes your photo livelier.

For the fun, here is the same serie derived in Instagram format.

Another exemple when you ask midjourney to provide you with variations on a single subject: 

It' s interesting to see how your brain understands they are the same thing.

Same compo, an old man, a bike, photorealistic, bokey, dusk.

By the way, it is possible to ask to describe picture to Midjourney. I uploaded the first photo and it provided me 4 prompts: 

1️⃣ dazzling cityscapes, grandparentcore, extremely gendered, reductionist form, solarizing master, night photography --ar 4:5 

2️⃣ a man holding a bike, in the style of grandparentcore, dazzling cityscapes, city portraits, soft-focus, stan lee, gerard van honthorst, portraits with soft lighting --ar 4:5 

3️⃣ man is old and in his 50s, in the style of dutch and flemish, street-savvy, bokeh, uhd image, evocative environmental portraits, illuminated visions, soft-focus portraits --ar 4:5 

4️⃣ a man standing on a street, in the style of grandparentcore, light-focused, dutch and flemish, flickering light, romantic emotivity, precisionist, chris claremont --ar 4:5

Of course you can endless play by generating a new image based on one of these prompts. Here is the results from the 2️⃣: 

Thank you for stopping by. What do you think about it? 
Let me know in the comments.