Using Midjourney for logo design
Can Midjourney AI generate logo design ideas and drafts?
☕ If you found this article interesting or helpful, you can buy me a coffee to support my caffeination needs: ko-fi.com/hortfrancis
Let’s get started.
I am new to using Midjourney, but have been using ChatGPT for a few weeks now.
I have previously worked as a graphic designer, and designed logos and branding for organisations.
The trick is to pass in a good ‘prompt’. Midjourney’s documentation describes a ‘prompt’ as:
A Prompt is a short text phrase that the Midjourney Bot interprets to produce an image. The Midjourney Bot breaks down the words and phrases in a prompt into smaller pieces, called tokens, that can be compared to its training data and then used to generate an image. A well-crafted prompt can help make unique and exciting images.
The trick to using AI well is to create good prompts. This discipline is known as ‘prompt engineering’. You can find out some more info on this here.
To try to get a really good result, I have used a technique called ‘meta-prompting’. I have used ChatGPT to help me generate the prompt.
Looking at the documentation for prompt writing with Midjourney, I don’t think this result is as effective as it could be. Midjourney states:
The Midjourney Bot does not understand grammar, sentence structure, or words like humans.
An approach that could improve the output of the production of prompts for Midjourney using ChatGPT would be to provide more context in a chat thread, pasting in more of the documentation and drawing attention to specific parts. Doing so would create a ‘continuous semantic context’, which would allow ChatGPT to provide more useful output over time.
I took the output from ChatGPT and edited it, then passed it to the Midjourney bot on Discord.
We get a few good drafts. Midjourney always makes four versions, and users can select one of the four to develop further. There are common themes here: the sun, with circular shapes and bright golden colours. The motif of the solar panels appears as a background in the top two.
Note the generic, nonsense characters: as Midjourney is averaging from billions of data points, it isn’t actually generating text data, but rather approximating the shape of text characters. We can see that it has gone for sans-serif in all-caps.
I like the no-nonsense utility of these designs, vaguely reminiscent of a military emblem. The NATO logo and Battlestar Galactica logo spring to mind.
Midjourney has somewhat of a ‘house style’, and the documentation mentions this:
Very short prompts will rely heavily on Midjourney’s default style, so a more descriptive prompt is better for a unique look.
Helpfully, Midjourney will keep generated images in your profile, so you don’t need to go hunting through the Discord thread, which gets new posts very quickly.
Rather than four logo drafts, I want a big collection of first draft ideas. My next prompt is: Many first drafts logo community company solar panel sustainable energy
We don’t get quite what we want, however:
We’ve lost what worked well and seemed appropriate from the previous attempt. Midjourney is ‘confused’; that is to say, the distance between what I want and what Midjourney has created is too large. Consider the concept of ‘theory of mind’. Effective prompt engineering involves adopting an effective theory of mind towards AI you are working with. We need to understand how it is likely to interpret words and syntax in a prompt.
I think the logo designs provided so far are too complicated. Here is an Instagram post I made about my rules for effective logo design:
A good logo should be simple enough that it will be recognisable at a very small size, and visually interesting enough that it will be memorable at a very large size.
To attempt something closer to what I might produce, I used the following prompt: Minimal Bauhaus logo solar panel community company energy sustainable
Graphic design associated with the Bauhaus art school uses a limited palette of bright colours, and simple geometric shapes.
Note that Midjourney doesn’t really understand what I mean by ‘Bauhaus’ here; is it the name of the company? The word itself appears in all of these designs.
We do certainly get some more interesting results, though. I particularly like the top-right version. I think it captures the idea of the sun, a modern building, and energy being focused towards development, in an open, uplifting way. This could be developed further.
The user interface for working with Midjourney via the bot on Discord provides options for developing any of the four images it produced.
Clicking V2 will give us variations on the top-right image.
We get some very slight variations — you might make something similar from iterating on 2nd drafts while designing a logo. This is an excellent colour scheme, particularly the off-white background.
A good workflow might be to take this image into Adobe Illustrator (or your vector graphics suite of choice) and make it into a series of vectors, to manipulate further.
What I did next was to use the existing images I liked as URLs in the prompt for another image.
Because Midjourney saves your images in your profile, on its website, you can copy the image link from here straight into the Discord chat. Using two of the images I liked as prompt data in this way, results in this image:
We’re getting a strong ‘solar punk’ aesthetic, which I like. Notice that we are getting more drafts per image: now 16 per image. This is closer to what I originally wanted, which was a sheet of lots of logo ideas. The circle shape absolutely dominates, so it’s hard to say if we are getting the minimal ‘Bauhaus’ logo coming through or not. Why don’t we take this new image, and add the Bauhaus design in as a URL image parameter, with this most recent image, and no words, to make sure?
Lots of interesting shapes here, easily made into vectors. There’s a certain ancient Egyptian vibe happening, and Midjourney seems very fixated now on circles. Fans of oranges will be impressed. Maybe we can take out circles and unlock some more creativity?
The documentation offers various parameters. I used: [image url] logo --no circle
I very different look. Much more colour. There are still circles, but the logo
keyword has given us what looks like a portfolio arrangement of small and large views of some work. It looks a lot like an averaging of the kind of thing you might find on Behance.
Conclusion
I think you can clearly use Midjourney to ‘brainstorm-without-a-brain’, to give you ideas for how to approach the task of visually communicating a set of ideas. More practice will be required to learn how to get more useful results from Midjourney, as it has a tendency for generic solutions and regurgitating design trends. The ability to feed images back in is exciting, and I would like to try this by passing in my own reference or 1st drafts, to see if new visual ideas come to the surface.
Meta-prompting with ChatGPT was not effective, because ChatGPT did not understand what Midjourney needs for an effective prompt. To improve this, it would be worth showing ChatGPT examples of what an effective prompt looks like.
☕ If you found this article interesting or helpful, you can buy me a coffee to support my caffeination needs: ko-fi.com/hortfrancis