🖌️ AI Art Tips Thursday #15: What's new in Midjourney v5.2 - Part 1

💡 AI Art Tips for the Day

Midjourney v5.2 is quite a huge improvement and I will go through all the updates in two issues so that I can cover them in enough detail for you.

Sorry for all the people who voted last week, I will explore the permutation techniques on the AI Art tips Thursday #17. P.S. I actually used this feature quite heavily to prepare for today’s materials.

For today, I will go through mainly the following:

• Base image creation quality: i.e. see what changes by just changing the midjourney version in the prompt

• Stylise parameter: in the announcement for v5.2, it is claimed that “a strong effect on the amount of the stylization” will be applied to the generation.

  

• /shorten prompt: finally, a prompt analyser of sorts that helps us to understand what unnecessary prompt words we can remove

Base image creation quality

To ensure the art generations among the different versions are based on the same foundation, I have set the same prompt and seed for the test.

First, let’s try an image with a human in the prompt 👇️ 

At a glance, I think both v5.1 and v5.2 performed equally well in this case. However, if you look a bit deeper, the generation from v5.2 has more of a different depth to the objects captured, i.e. the ceiling light shown on U4 of v5.2. So yeah, some improvement can be seen from this new version.

Now, let’s try a more oversized item 👇️ 

In my opinion, v5.2 wins this one for sure. Not only the quality of the individual image is way better with more detail and depth, but also I find the variation of the different styles among the 4 outputs to be much larger. This allows me to explore different art directions further, which is one of the main reasons why I love AI art generation so much.

Lastly, let’s get micro 👇️ 

Apart from the details from v5.2, like the reflection on the water, I think v5.2 and v5.1 perform almost the same. It is really up to the individual’s taste to find out what he or she likes.

Overall speaking, I think the base image creation has definitely seen some improvement from v5.2. The main improvement I can see is the additional details and depths added to the image to make them more vivid and lively.

Stylise parameter

Similar to the test of the base image generation, to ensure the art generations among the different versions are based on the same foundation, I have set the same prompt and seed for the test. For the test, I have tried v5.1 and v5.2 to see if there are any differences in terms of how the stylise parameter performs.

First, let’s see how the v5.1 works 👇️ 

For v5.1, it looks like the variation from the different stylise parameters is not that high. To be honest, if it is not because of the two horns for U1 in the --s 500 option, I thought I have pasted the wrong images there. The colour palette and the composition are quite similar across all 4 variations of the stylise parameter.

In comparison with v5.1, the art generation from v5.2 using the stylise parameters differs a lot in my opinion. Other than the main object being changed to a standing unicorn-looking man, the general colours and composition change a lot as well. Although, I must say the output from --s 750 and --s 1000 are still quite similar.

Nonetheless, overall speaking, I think v5.2 certainly achieved the goal of getting a more stylised generation using a higher stylise parameter.

/shorten prompt

Now this is something I didn’t think midjourney would create, but it is certainly a good feature to be used.

Let’s take a real example, the following is an image I found online that is claimed to be created from midjourney v5.2

The prompt for this image is the following:

white tulip bouquet opening, liquid smoke fractals, elegant botany, dynamic movement, raphael, caravaggio, surrealism, dreamy, hauntingly beautiful, moody --c 25 --ar 2:3 --s 400

Well, it is certainly quite lengthy and you might have the question, do I need to type all the words in the prompt to get something similar to the above image? Luckily, /shorten is here to help you out. Here is what I got by putting this prompt to the /shorten prompt:

The words you see that are crossed are words that do not add any particular value to midjourney, i.e. unnecessary filler words. The ones that are in bold are the ones that would influence the prompt results more. Also, you may notice that the parameters are removed from the result window. I assume that those parameters, e.g. --c 25, would always have some impact, hence they haven’t been included in the results shown.

If you click on the “Show Details” button, you would see the actual numerical value of impact on the prompting for each word.

Now to prove the shortened prompts actually work, I have tried to generate the same thing art using the shortened prompts and look at the results below 👇️ 

Pretty close in my mind, I know the actual colour of the tulips is a bit different, but the composition and the details are quite similar. So in summary, you can now clean up your prompts to achieve the same results in art generation with fewer words.

Alright, that is all I want to cover for today. I know I have skipped the big thing, i.e. Zoom out feature, for today. The main reason is that I want to do more exploration so that I can show you what you can do more with the whole zoom-out feature family. I hope you like what is covered today.

🖼️ Image of the day

Personally, I really like this unicorn-man-walking-on-the-street image. This is also quite similar to the previous images I have seen with different types of animal and man hybrids, mostly in some sort of working environment.

I will use the following image as my foundation for the zoom-out feature, so stay tuned!