Table of contents
- Logo generation
The latest cool thing we’ve all heard about is the DALL-E 2 (hereafter Dall-e). When the first previews began a few months ago, this AI model was basically ubiquitous . A few weeks ago, the doors opened to using the model, giving access to many people on the waiting list. I was one of the people who got access.
Once I got access to Dall-e, I spent the day learning a few basics (e.g. adding “artstation” to the end of a phrase automatically made the output much better). In fact, it generated a lot of (some of them nice) images. So, I’ve already done some form-up.
A little more background: OctoSQL , an open source project I’m working on, is a CLI query tool that allows you to query multiple databases and file formats with a single SQL query. I had always thought that the logo should be renewed, but with the advent of Dall-e, I was able to combine the fun (of creating a logo) with practicality.
In practice, you’ll find that it’s a bit like the storytelling process of the writers depicted in Season 4 of “Westworld.” , but I recommend watching this drama series if you haven’t seen it).
So here is the final logo.
In the continuation of this article, you can see where I started, what I went through, what I learned, and how I slowly evolved into the final image I chose. In the following, I will only show you a journey that I have been quite satisfied with. In other words, I will only show the images that were reasonably good (the remaining 70% or more were terrible and I got rid of them).
Before that, let me briefly explain how the DALL-E 2 works. Users get a set of free credits to start with, and can purchase more. For one credit, you can do any of the following:
- Generate: Generate 4 images for the specified phrase.
- Edit: Generate 3 images according to the given phrase for the transparent marked area of the original image (passed by the user as input to DALL-E 2) (image editing software or built-in transparency drawing tool).
- Variations: Based on a given image, generate 3 variations without specifying a phrase . In other words, it is unfortunately not possible to do something like “xyz (where xyz is some phrase) while keeping the image itself the same”. However, variations can be generated for the transparently marked areas mentioned in point 2.
I had a pretty specific idea for a logo (I thought I had it, but apparently couldn’t put it into words). The name OctoSQL comes from octopus and SQL. The idea is that OctoSQL has many arms and can manipulate many entities at the same time, while OctoSQL can manipulate many data sources at the same time.
So the first thing I wanted to do was a cute cartoon octopus that manipulates a bunch of databases (or entities that represent databases, I decided not to use actual database logos).
Let’s start with easy phrases. You can see that the following words, which I typed quickly, use keywords that are “directional” suffixes (prefixes).
The first one (left image) looks really cool. So let’s print some variations.Pretty good! It looks cartoonish, although it would be nice to have a little more depth. But the biggest problem is that this “octopi” (the beautiful plural of “Octopus”) has charts. OctoSQL isn’t dealing with charts, it’s dealing with data. In these images he may give false expectations about what is possible with OctoSQL.
Now let’s get back to drawing (or rather, phrasing) drawing. Let’s add some abstract shapes that the octopus should have.
Not what I really want, but it still feels pretty cool. But I think you can experiment with drawing styles (using the image above).
A baby octopus juggling 3d shapes representing databases, arms wrapped around one cube, data flow through the cube, digital art, cartoon, line drawing, logo, simple shapes arm wrapped around one cube, streams of data passing through the cubes, digital art, cartoons, drawings, logos, simple shapes)
The above output image is disqualified because it is a simple figure.
Let’s try some more.
Now, we can go back to our original approach, but could it be simpler? Use shapes instead of databases and add tags to improve quality.
What about the quality of the images above? not bad. how about a logo? Not so good.
Let’s try sketching with a pencil again.
It feels good! But this variation doesn’t produce anything breathtaking.
How about using something more abstract? data flow. And what about adding a “logo” to the input?
The above images are funny and worth generating. But it doesn’t feel right to use as a logo. Maybe I’m asking too much of Dall-e. Try a simple octopus with no databit description.
The above images are very attractive! Let’s edit these and add various things to that arm! For example, add data streams, data records, 3D shapes, and the blue data storage cylinders that you’ve seen on all the blueprints you just learned the name for.
Each time I generated an image, I added a (transparency) mark to replace a part of it, and then let Dall-e do the work.
The above image was not bad, but I decided to output it again by adding only the “logo” tag without inputting unnecessary qualifying words.
I had an epiphany by typing the above. Logomarks often have a background. This dark background circle was exactly what I needed. I also found that Dall-e’s output was forced to fit inside this circle (so I could avoid drawing in all the empty space (outside the circle)).
I tried out some basic phrases and soon had a nice logo.
Baby octopus, logo, digital art, drawing, in a dark circle as the background
…and the octopus sits neatly in the central space, making it a very useful logo.
Also, it might be a good idea to add an entity that the octopus can play with.
However, the above image turned into a dangerous-looking swarm of octopuses. I had already used the word “baby” to make it a cute image.
So let’s add a cuter element (different from “baby”).
Cute baby octopus playing with cubes, logo, digital art,drawing,in a dark circle as the background.
Much better! Can’t we add more cute elements? Shouldn’t it be!
Cute baby octopus playing with cubes , logo, digital art, drawing, in a dark circle as the background, vibrant , cheerful, bubbles)
Chaining variations based on the prompt input sentences above resulted in a nice octopus, a bad octopus, and a bunch of app icons.The output above isn’t really going in the right direction yet. Then maybe we can edit the original input sentence to suit our needs. So far the output image is very playful. But OctoSQL is for serious data processing, and the logo should reflect that app’s capabilities. And to be honest, I couldn’t think of anything more trustworthy than a yellow safety helmet (which, to be honest, is clearly a better logo), so let’s add that.
I dragged a transparent brush over the top of the octopus and gave it the phrase:
Cute baby octopus playing with cubes wearing a yellow safety helmet, logo, digital art, drawing, in a dark circle as the background, vibrant, cheerful, bubbles)
The output image above looks really nice.
I’ve taken a little detour here to generate various octopuses with helmets (and variations based on them). As a result, it didn’t do anything useful, but I was happy with the cuteness of the finished octopus, so I’ll leave it here for readers to enjoy. Thus, more plain octopuses playing with cubes were created. I didn’t use these for the logo, but they look great.It occurred to me that the cubes, data records, databases, and blue data storage cylinders I’d created above might not be things that octopuses are accustomed to playing with. What about something more natural, such as mussels?
Cute baby octopus playing with mussels, logo, digital art, drawing, in a dark circle as the background, vibrant , cheerful, bubbles)
This output really struck me. It’s a good size, it’s cute, it’s almost like a real octopus (not much artificial stuff), and the mussels look nice.
So I added a helmet and also edited the sides of the helmet to fix the artifacts (image noise).
Cute baby octopus playing with mussels wearing a yellow safety helmet, logo, digital art , drawing, in a dark circle as the background, vibrant, cheerful, bubbles)
The octopus’ left eye was also broken. Only his left eye was modified with a transparency brush, leaving the input phrase the same. By doing this, Dall-e regenerates only his eyes.
Fixed! Now back to the mussels. They look cute, but from a distance they are too many and too small. Moreover, this does not convey the original message (the logo of the SQL execution application). So, let’s draw tentacles and add “play with 3D figures” to replace the mussel with some kind of “3D figure”.
I really like the first one, but the rightmost sphere is too big. Besides, it might be possible to add a figure other than a sphere. So let’s paint on this sphere and see if we can get something nice like “3D figures like cubes, spheres, etc.”That’s how the center image you saw at the beginning of the article was adopted. The sphere inside the cube is interesting, the octopus is manipulating some objects, it’s logo-ish, it’s pretty cute, it’s (somewhat) believable, and it doesn’t have small elements. It looks good even from a distance.
After outputting the logo image, use Procreate to make the background transparent, then make the image a little smaller so that the bubbles surrounding the octopus take up the entire screen. became.
Finally, just to be sure, I did a lot of reverse image searches on Google (for similar logos).
To be honest, I would have preferred a simpler image without the complex shapes, but I couldn’t convince Dall-e to do it that way. That being said, I am very happy with this logo.
Overall, making a logo with Dall-e was a fun journey. The final result wasn’t ideal, but I’m very happy with it. As for Dall-e, I think it’s still mostly in the “first iteration” stage (the main exception being the pencil sketches, which are surprisingly well done). In my opinion, artists have nothing to worry about right now (kind of like GitHub Copilot). On the other hand, stock photo websites should be very concerned about the rise of Dall-e…