Logojoy was designed to replace myself as a logo designer. To achieve that goal, artificial intelligence is necessary. First, we need to talk to the customer the way a designer would – by gathering inspiration and assessing brand values. Second, we need to continue to iterate based on the customers preferences. Once Logojoy starts to learn, it can actually become better than a designer.
Logojoy applies AI by learning each customers preferred logos and then producing better logos based on previous customers logos. For example, Logojoy is able to learn that Helvetica goes really nicely with a more ‘geometric’ icon, just like a designer would. Another example would be learning that thinner fonts need a darker color to stand out.
Logojoy uses AI on a universal level across all of it’s customers over time, we use non-AI algorithms to learn each users preference in logos.
Step 1: Assessing preference
Logojoy learns via AI by first tracking two things: the customers ‘favorited’ logos, and the changes that they make on each logo.
Customers are able to ‘favorite’ logos and we track each time they do this. They are also able to make changes to the logos – such as changing fonts, colors, icons, etc. We track every time they make a change and more importantly, we track the difference between the original favorited logo and the new, modified version. By tracking this difference (usually between 10-20 changes), Logojoy starts to learn and see trends across our entire userbase.
Does everyone change the font to a heavier font when the logo has a light color? Does no one like Helvetica combined with a floral graphic? Do longer company names look better with a less decorative font?
Step 2: Applying learnings
As Logojoy learns, it’s able to apply it’s learnings instantly. As every new logo is generated, our design algorithm heavily communicates with our AI API to make sure it’s creating a great logo.
Step 3: Genetic Algorithm
The last step isn’t actually AI – it’s a genetic algorithm. As you ‘favorite’ logos, we apply a natural selection type process to generate more accurate logos as you scroll. Each logo is made up of hundreds of ingredients (ie ‘funky font’, ‘bold color’, ‘logo has a symbol’, etc.) – these are called traits in the context of genetic algorithms.
As you favorite logos, Logojoy takes the individual traits and adds them to the gene pool with a score. The more you pick logos with certain traits, the higher that traits score is. For example, if 6 out of 10 of the logos you pick have a sans-serif font, there will be a 60% chance the generator will use a sans-serif font when generating subsequent logos.
By using a genetic algorithm, we’re able to weed out poor designs and bring the best designs to the top, just like we see in natural selection.