A little bit of a bonus post this time! As well as being interested in travel and photography, I am quite a bit of a nerd. That might go without saying as it is VR photography that interests me, and I coded my own VR gallery…
Anyway! Keen to leap aboard the latest fad, I read an article about a newly released AI art generation model capable of producing images comparable to DALL-E, or Midjourney that can be run locally on your own computer and decided to give it a go to try and create a site logo. I have tried local AI art generation before, as well as using notebooks on Google Collab, but with mixed results. It certainly made the long journey to the US more interesting last winter, feeding the model photos like one of a plane from the airport and getting it to generate images such as the one below using prompts like “hell demon emerges from plane”:
The idea that artificial intelligence was creating the image fascinated me, but the results were still more abstract than realistic. The article I read pointed to the following Github repository: https://github.com/lstein/stable-diffusion
While I struggled getting the suggested Jupyter notebook to work, I was able to get the script running using the Anaconda3 method within about 20 minutes – and the results were pretty encouraging! Using prompts such as “scottish highland moody sky” the AI model was able to generate photorealistic settings that wouldn’t be out of place here!
This was not what I intended to use the model for, and while “baby elephant tucked up in bed” was cute…
I was looking for a logo! I started by feeding the script a very basic prompt to see what it came up with – “virtual reality logo”, and while this image would make a powerful logo for a ruling corporation in the year 2150, it did not match the feeling I was looking for.
My favourite of these images was the following:
While I love this idea for representing virtual reality – the 2 opposing colours merging in an eye – it does not really convey a feeling of nature, or relaxation, and it’s links with VR are probably way too abstract!
After sleeping on it, my final idea was to try and create a fairly simplistic drawing of nature. Something like “a childs drawing of a mountain”, “simple line drawing of a mountain”, and finally “ink drawing of a mountain”. The results for these prompts were far more promising – and it is also possible to feed an image to the art generation script to essentially iterate on an image that is close to what you would like. Following this process of generation, and iteration, I landed with a few promising images:
All of these had qualities that I liked, and their simplistic natures would lend them to becoming logos far more easily, perhaps with a tweak here or there. In the end I settled on the image used in my current logo, a simple drawing of a hilly valley. I added text over the top – and I have to say, I am incredibly pleased with the end result. All in all, I feel like I have ended up with a classy logo that matches the feeling of what I am trying to create, aided by a free AI art generation tool.
