Aagje Reynders
Previously, I wrote a blog about how AI generation works and about the controversy surrounding AI as an art form. Today, I will demonstrate the creative process behind generating images and how you can use tools like Midjourney and Magnific to bring your idea from concept to finished product.
For this blog, I wanted to finally create some regal Renaissance portraits of my cats. I attempted this last year without success, possibly because my prompting skills weren’t there yet. In addition, I wanted to try again because Midjourney has since added some cool improved features like the ability to use reference images.
If you’re interested, I’ll guide you step-by-step on how I created these cool portraits of my cats!
Before we start generating images, allow me to introduce the subjects we’ll be working with. I have two purebred cats: one Cornish Rex named Luna, and the other, a Siamese, who listens to the name Percy.
I mention their breeds because it will be quite challenging to create these portraits to actually look like them. On one side, these breeds are pretty rare—there aren’t many pictures of Cornish Rexes around, and they often get mixed up with Sphynx cats. On the flip side, ‘Siamese’ is both a breed and a color pattern any cat can have, making it tricky for the AI to nail down the breed characteristics. Nevertheless, let’s explore various methods to overcome these challenges.
As you can see here, without adding any reference images, the results are pretty far off from how my cats actually look. While the first image of the gray Cornish Rex actually bears a lot of resemblance to Luna, the third image in the series already looks more like a Sphynx.
Rest assured, I won’t burden you with a detailed explanation of the breeds. Instead, I’ll focus on the solution. With Midjourney, you can add reference pictures in three different ways: character reference, style reference, and image prompt.
💡 What is the difference between character ref and image prompt?
Here’s a practical example to illustrate the difference between the two. I accidentally used a reference image of Luna peeking from behind the curtain as an image prompt. You can see that the generated cat doesn’t look like her; it’s merely following my prompt. However, the cat’s posture and the curtain in front of it closely mirror the original photo. This exemplifies how an image prompt excels in capturing composition and poses, while not reproducing the original image in its entirety.
In this case, we will use ‘character reference’ to make sure it looks like my beloved cats and ‘style ref’ to make sure it looks like a Renaissance oil painting.
Prompt: A highly detailed Renaissance-style portrait of a Cornish Rex cat. The cat has large, pointed ears, striking green eyes, and a sleek, light gray curled coat. It is depicted wearing intricate lace ruffs and ornate jewelry, including pearls and precious stones. give her lots of lace ruffs around the neck. and earings in the ears. The background is dark, adding contrast and highlighting the regal attire and dignified posture of the cat, similar to a noble figure from a Renaissance painting. Make it look like an oil painting.
Here’s where my process began. I selected a photo of Luna as the character reference, and three images of regal-looking ladies as style references.
It wasn’t completely off the mark, but it was clear that the system was struggling to balance the style, the character, and my prompt. It was falling short on all fronts. So, I took a step back, and focused solely on transforming Luna into an oil painting.
💡 Wonder how I wrote my prompt?
Midjourney offers this neat feature where you can upload an image, and it provides a detailed description of its characteristics. I basically use this as my starting point to get the exact terminology that describe what I want. It’s especially useful when targeting a specific style that you can’t quite name or describe. When prompting for photography, this feature can assist you in selecting the appropiate lens and settings to get the intended mood.
Prompt: A highly detailed Renaissance-style portrait of a Cornish Rex cat. The cat has large, pointing ears, striking green eyes, her fur is gray and curled. Give the cat a curly furred fur. The background is dark. Make it look like an oil painting.
Notice how detailed the cats are now? I really liked the first image, so I started refining it further using the repaint option.
Bear in mind, every phase is an iteration. I might show you a screenshot of the result, but in reality, I may have tweaked the same prompt multiple times with small adjustments. Prompting is all about trial and error, and you need to try lots of things to get the image you’re aiming for.
Going back to the repainting process, I’ll basically select the areas that need to be regenerated, and add a new prompt. Let’s see how the images turn out:
Pretty good, right? It’s worth mentioning that I’m no historian, so if you want historical accuracy, you’ll need to do some research on what royals actually wore. I’m not aiming for accuracy—I’m aiming for an iconic look for my cats.
Using the repaint tool, I experimented with various headpieces, styles, jewelry, ruffles, and more, repeatedly, to get the royal look I wanted. I also removed some odd-looking ruffles (like the ones in the back).
After roughly 20 iterations, I ended up with this, using just the Midjourney tools I described earlier:
💡 Reframing:
Another tool I love is the ‘reframe tool’. It allows you to adjust the size or to zoom out, and it will automatically generate additional content—which is great for improving composition. For example, here’s a zoomed-out view of Luna, which reveals more of her dress and jewelry:
Personally, I’d probably start over from scratch and create multiple generations to give myself a few options to choose from. For the purposes of this blog, I don’t have time for that, but if I were working with a client, I’d strongly recommend it.
If you’re striving for pixel-perfect accuracy, I’d also suggest making some tweaks in Photoshop. Midjourney is a great starting point, but it doesn’t have to be the final step in your creative process. For example, I’d remove the neck folds and add some fur texture. I tried doing that in Midjourney, but it wasn’t really cooperating, so Photoshop would be more effective—For time’s sake, again, I skipped that step here.
Next, let’s move on to the next phase: using Magnific. It’s an excellent tool that upscales your images and adds a lot of detail. You can enter a prompt and adjust the settings, but I generally prefer to keep things relatively simple.
For the prompt in Magnific, you have multiple options at your disposal. You can re-enter the same prompt you used in Midjourney to help fine-tune the details. In this case, I really wanted to retain that painting effect. You can assign numerical weights to each keyword in your prompt, ranging from 1 to 1.4, to indicate its importance in the composition.
I also experimented with adding fur, which was successful to some extent: it unintentionally made some fabrics look furry as well, so I decided to leave that part out.
Here’s the final result:
For Percy, I did exactly the same:
In conclusion, the process of crafting regal Renaissance portraits of my cats using AI tools like Midjourney and Magnific was a blast! By playing around with different prompts, reference images, and tools, I was able to create unique and intricate images that truly capture the essence of my cats. Midjourney was a solid starting point, and upscaling with Magnific added those final touches to perfect the overall look I wanted.
These AI tools are incredibly powerful, but they’re merely the first step in your creative process. Don’t be afraid to tweak, refine, and improve your images to get them just right. Whether you’re a hobbyist or working with clients, embracing an iterative approach and combining multiple tools will allow you to create stunning, personalized images.
Written by
Aagje Reynders
Want to know more?
