Gain Control Over AI-Generated Images: Art Terminology vs. Artist Names
Are you struggling with gaining control over your Midjourney outputs? This quick guide might be useful. I also have encountered this problem and I’ve been gradually tweaking my prompts to produce outputs that match my vision.
Lack of control is something I’ve noticed in many mid-journey generations. A lot of the most “artistic” results I see on the trending and hot community tabs on the site are often the result of happy accidents. For instance, I’ve seen users prompting Midjourney to generate comics in the style of Kelly Sue DeConnick, however Kelly is a writer, not an illustrator. I figured that the image model knows the kind of comics Kelly writes for and can pull data related to her as a writer. And while happy accidents can create beautiful images it also means a lack of control over the output. Some of these images are also not usable for specific cases, such as image references, publications, games, packaging, etc. If you need consistency across your output, you need to keep certain words aka variables the same.
Think of image generation as coding. You have variables, constants and parameters.
- Constants are words that will give you that consistency.
- Variables are words that will modify the subject.
- Parameters such as aspect ratio, stylize, chaos, and version will modify the quality of the output.
Also, I will like to add that I consider prompts inputs and images outputs. It’s a type of computer lingo that I like to use for this image generation process.
Limitations of using artists as prompt properties
Many people rely on using artist names as a way to gain more control over the output of their image generations, or they combine several artist names to create new styles. However, this often leads to a standard look in these images, particularly in the painting and illustration categories. There’s nothing inherently wrong with this, but I also want to achieve interesting results without depending on contemporary artists as properties in a prompt..
Examples:
Brooke from That 70's Show, most beautiful, visible abs, crop top, full body, intricate, cinematic lighting, art style by artgerm --v 4 --ar 9:16
The Dairy Queen, a white-haired woman with long white hair, a crown on her head and a crystal dress with white feathers against a background of blizzard, 8k, dynamic pose, character design, frost colors, white, josephine wall, robotic, wonderland, marvel, sanrio, priestess, ♀️, ❄️, 🌸, 💗, 💻, balletcore outfit, frontal lighting, soft lighting --ar 2:3 --chaos 20 --iw 0 --s 750 --q 2 --v 5
Gaining control using art terminology as prompt properties
I’ve discovered that incorporating contrasting descriptions of art mediums, techniques, subjects, and art movements in prompts can produce similarly intriguing results as using artist names as references. This can be a valuable tool for prompt crafters who have more knowledge of art history and techniques and want more control over their output.
Examples:
Digital oil painting, abstract expressionism, cute silver witch riding a huge black cat into celestial battle, expressive and impulsive brushwork --ar 3:2 --s 1000 --v 5
Impressionist oil painting, 16th-century Spanish Buccaneer resembling Pedro Pascal, on a ship in the Caribbean Sea, wide-shot seen from above, fighting pose, warm colors, blue hour, cinematic, impulsive and expressive small brushwork --ar 3:2 --v 4The images below take the impressionist art movement and combines them with a cyberpunk theme to create a more contemporary look. Both images use both words as constants and variables.
Impressionist oil painting, Young Cyberpunk man wearing cybergear, on rooftop with a background of a futurist city. Wide-shot seen from above, warm colors, blue hour, cinematic, impulsive and expressive small brushwork --ar 2:3 --v 5
Impressionist oil painting, Young Cyberpunk man wearing cybergear, standing on the rooftop with a background of a futurist city. wide-shot seen from above. Vibrant Neon colors with color doge effects, blue hour, cinematic, impulsive and expressive small brushwork. Inspired by Cyberpunk 2077 --ar 2:3 --v 5Side by side comparison Artist Prompt and Technique/medium Prompt
cute anime cartoon girl, round face, button nose, short dark brown hair, red jacket, round glasses, blue beret, by James Jean --v 5
gouache illustration comic of a cute anime cartoon girl, round face, button nose, short dark brown hair, red jacket, round glasses, blue beret --v 5You will noticed a few things in the two images above. The one using the artist reference doesn’t have the correct hat, nor it fully follows the description of the shape of the face I specified since it’s introducing the bias of the artist’ style. I’ve explained how artists’ style works in Midjourney in this [ article ] where I compare artists’ artwork with Midjourney model’s interpretation of that style.
Why is this important?
Primarily, using artists as references gives you less control over the output, as the images generated are more biased towards the artist’s shape language, color palettes, and themes. If you want to have more control over your generated images, you should consider not using artists as references and instead focus on art terminology and techniques.
Conclusion
This is just the first part of a guide on how to create characters using Midjourney. In the next part I will expand on how you can craft prompts that will give you outputs that are closer to what you have in mind. Language can be tricky, particularly for non-English speakers, therefore I hope that with this series of articles, I can demystify and explain how to craft better prompts using simple language.