Using AI to Generate Icons

Since I have minimal design skills, I went back to Fiverr (the digital gig economy platform) to get some icons done for CodeTrimmer - explaining that I wanted something like a “pair of scissors floating over some computer code”. At the same time I’ve been playing with DiffusionBee - a free Apple silicon version of the Stable Diffusion artifical intellligence that generates images from text prompts. The image above was created on an M1 Macbook using DiffusionBee.

Since generating an image from a text prompt is exactly what I’d just done with Fiverr, I thought I’d give the AI a chance. Even if the results are not great, they will do for the toy applications I’m creating as I’m learning.

The model in Stable Diffusion is trained on images scraped from the internet, so it’s strong on pop culture and painting styles. For instance “Jack Black as Aquaman” is likely to give good results, also “Kermit dressed as Yoda in front of a sunset landscape”. The license for the outputs of the AI is quite permissive, including for commercial purposes as long as it’s not used for evil (that’s my paraphrasing, the actual CreativeML Open RAIL-M license is here ).

Here’s a couple of outputs for the different versions of the “scissors floating over code, macOS, icon” prompt.

Like any tool, there is a skill to using it well, and with image generation from text prompts, the skill is in creating the prompts, hence a sudden plethora of articles about this all over the internet, and even a market for them. So perhaps with a bit more skill I’d get something closer to what I’d like, although if you are prepared to accept something a bit abstract a couple of these would be fine.

If I had a small amount of design skills these images might serve as a great creative thinking starting point - for example I like the criss-crossy background of one to represent code. I like the drop shadow and flat green of the ipod shuffle one - so these ideas could be combined to make a pair of flat green handled scissors floating over the lined background.

If you are not allowed to use AI’s for evil (I’m looking at you Skynet ), it seems like using them as a tool for creating okay images for apps and other digital content is likely to become a practical use for them.