Originally written on 23rd June 2024
Playing with AI models (part 1)
I spent the last week playing with Tensor AI, a free image generating AI website. I was made aware of this through an AI artist whom I interviewed. They spoke on the subject of not having digital art skills but having the desire to create works of their favourite characters through the use of Gen AI tools. My earlier assumptions were that a generative model of AI required text input that were descriptive and would generate a close enough image to what is desired in a “generic” AI style. My experience this week has taught me that AI generated art is harder than it first appears, especially if you are trying to achieve a specific output. Tensor AI had a large number of base models that have been developed to cater to different styles and different characteristics models. It is a large part of model training that has been done by individuals or groups and can be used to build up from. To generate images, one has to build upon these models and spend time experimenting on how different text prompts translate into images and how different prompts interact with one another. I see this field of art evolving into a specialised field, similar to the use of digital design software’s such as Adobe or Corel. While it makes the creation of images faster, the training that is required for individual creators does take time, interest and engagement. It also requires research to create different styles of work. The same prompts may translate very differently on different models. If you spend a lot of time getting comfortable with the use of one, it may be difficult to shift to a new base model. I therefore, see artists specialising in a unique style that they have developed through the combination of the generative models that they have chosen and expertise with the use of prompts learnt through experience.
Playing with AI models (part 2)
This week has made me experience something new. I personally may not shift entirely to using generative AI tools for my creative practice, but I can see it’s appeal. I was able to create a vast amount of image option, even though they were not meant for any end use. The anticipation of each new image being generated; what was going to pop up? An entirely unique and new visual piece was exciting, almost titillating. There was a sense that this was not an image I found on the internet but created using a software and will not be seen by anyone else. A private work of art (using that term loosely) that was spat out by the technology god, only for my eyes. I can imagine that people who create AI images must be experiencing a similar sense of purpose as they sit for hours testing prompts and perfecting a range of images. Though works of art created by AI cannot be copyrighted, they must feel a sense of ownership over their work. Many of these creators may have felt that they missed the ship on the digital art movement, but now have their own little boat to ride upon behind it.