Can Midjourney substitute a photo-shooting?

Federico July 28, 2023

While exploring the world of AI generative tools, one of the most famous is Midjourney.
What makes it so special?

It was one of the first;
It works and operate in Discord;
It generates incredible results.

In a recent project I was in need to use some images, but did not had the resources for a proper photo shooting as well as the time for it. Being low budget and short in time, I decided to play a bit more with Midjourney.

I play also with other tools like Stable Diffusion and Dall-e, and I will create a proper post to compare them, but I do like a lot Midjourney.

While playing with the tool, I had some decent results, and it took me really a small amount of time to get it done. But... not everything was perfect.

Let's see the results...

Midjourney experiments
Conclusion
- Some of other outcomes

Midjourney experiments

Here the list of prompts and results.

First run

Prompt: photorealistic image evening aperitif time frustrated girl with red hairs, rised right arm while sitting at a table with friends, in a bar in Luxembourg, looking at the waiter in the background which is distracted. The bar is at peak time and in the background a waiter is passing looking somewhere else. --ar 7:4 --c 10 --style raw --s 250

Not bad, but not quite near what I need and not really corresponding the description. A lot is missing.

Second run

What if I use a picture as reference to provide some more hints to the scene? My goal is to get the image I want, as much as possible.

Prompt: <image url> frustrated girl with red hairs calling waiter --ar 7:4 --c 10 --style raw --s 250

While the scene is matching the prompt picture, the girl is not really aligned with my expectations.

Third run

I am not really happy, so I decided to get back to full text prompt. With a shorter description.

Prompt: bored red hair girl sitting at a table with friends in a crowded bar, trying to call the waiter --ar 7:4 --c 20 --style raw --s 250

I had a nice result, but still a lot of the prompt inputs are not taken into consideration. Probably it is a limit of Midjourney at this time. I do like though the V4 here so I keep it.

Fourth Run

In my effort to get the most usable pictures, for my project, I wanted to run again the image prompt, to see if a second generation can be better. I changed one keyword in the prompt.

Prompt: <image url> bored girl with red hairs calling waiter --ar 7:4 --c 10 --style raw --s 250

Oh my, this time I think it nailed it. The V3 here is my choice to go.

Conclusion

Midjourney is and incredible Ai generative tool and it can produce impressive results. There are some tricks to learn in how to speak with the AI, how to build the prompt and this can be learned only by experimenting.

There is one big limitation only, it is not able to understand and comprehend all the inputs in the prompt, probably there might be some formats that can create better outcomes, but I have still not found the right balance.

Overall it is incredible and combined with Adobe Firefly and other AI tools, plus some editing, it can speed up a lot the work, at least in the explorative and concept phase.