Building a Vector Illustration Design System with Midjourney
For the last few months I’ve felt that it must be possible to create an entire library of illustrations in Midjourney. I’d ensure their was consistent styling and then I’d convert them to vectors with vectorizer.ai or a similar tool. However, that’s the sort of project that takes a long time and never gets off the back burner. So when Midjourney launched it’s style tuner feature last week, I knew it was time to act. I accepted that I didn’t have time to write the prompts for hundreds of images, so I settled for proof of concept with 16. I imagined that the brief was for a major outdoor retailer and I was aiming for consistency in style.
To use Midjourney’s Style tuner you simply type /tune where you would normally write /imagine and you continue to write your prompt as you normally would. I wrote: rough sketch, vector illustration, simple organic lines, spare, minimal, elegant, black, white, light brown, light green. Midjourney will then ask if you if you want 16, 32, 64, or 128 variations. Trust me, 16 is fine. I also chose ‘default’ over ‘RAW’ mode, but I don’t think it matters much. Midjourney will then ask if your sure, and explains that this will cost you a portion if your fast hours.
Once you click ‘Submit’, you will wait for a minute or two and then be directed out of discord to a web based URL where you will see 32 sets of images. You can either view these as a grid or as pairs arranged in two columns. The idea is that you will then pick up to 16 (minimum 5–10) and this will be the aggregated into into your final style. You are then provided with an alphanumeric code that you copy ad bring back to discord and append to your prompts with “— style 6neukDUZauN6”. Theoretically, all your prompts will now produce images in an identical style. If you don’t like the result you can return to the URL, modify your selections, and receive a different code. You can also share this code with other members of your team so you can all produce in the same style. You can even share the results of your tuning, so go ahead and click here to check it out. You should be able to view the results of my tuning and use my generated style.
In reality, I found that there was a still a little too much variation in the images that Midjournry spit out. In each group of four, I could always find one that matched the look I was going for, but I had to re-roll more frequently than I expected. Also, for most of my images I added ‘transparent background’ in the second round so I could easily isolate my subject from the background. If you want a consistent aspect ratio (like 16:9) it’s easier to add it to your original /tune so you don’t need to add it to every subsequent generation.
After I successfully had produced about 16 images in Midjourney, I downloaded them to a folder on my desktop and gave them a recognizable filename (such as hiker_1.png) so I could easily tell them apart. I then quickly converted them into vectors using vectorizer.ai. This is a free tool and this was the easiest part of the whole process. Simply drag your .png and 15 or 20 seconds later, download your .svg file. These were easily imported into both Illustrator and Figma. Every file has individually editable vector points and behaves as would expect it to. Except in Illustrator, all the paths are automatically grouped, so you can pick up the entire drawing and move it. If Figma, all the paths are ungrouped, so you will need to manually group them.
Overall I’m very happy with the results. I made 4 or 5 backgrounds in a 21:9 aspect ratio, several characters, and background and foreground objects. It was very easy to assemble them into scenes, storyboards, or concept art. With more time, I would generate a larger library and try to make consistent characters in a variety of poses. I could see how this initial workflow would be really helpful for moving into a graphic novel or motion graphics. You can view the Figma file here.
Outtakes and Bloopers
While the whole tuning system is meant to provide a consistent style, I guess Midjourney feels like taking some liberties.
