AI image generators have taken the creative world by storm over the past year, unleashing a wave of stunning surreal images created by text prompts. But some topics can still be a challenge for their algorithms. And it seems that includes cycling.
The site CyclingTips had observed the furore around the new technology and decided to put it to the test, by having several AI models generate images of scenes related to cycling. Close your eyes now if you want to avoid a horror show. Some of the results resemble the aftermath of a particularly gruesome Tour de France pile-up (to find out how the technology works, see our guide to how to use DALL-E 2).
Cycling tips (opens in a new tab) was intrigued by the impressive AI-generated imagery that’s been popping up since text-to-image generation took a leap forward with beta releases of diffusion models like DALL-E 2, Stable Diffusion, and Midjourney. He had to try the technology for himself.
“What would these AI platforms create if we fed a bunch of cycling-related prompts? And would any of them be suitable for use on CyclingTips?” he wondered. He used several AI models, including Steady broadcast (opens in a new tab) and the free browser-based app Crayon (opens in a new tab)and wrote text prompts asking for images of different aspects of cycling, from “a cyclist climbing a mountain” to “a showroom full of bikes”.
He found the results to be “fascinating, entertaining and, in some cases, more than a little horrifying”. Yes, it seems that AI image generators have even more trouble cycling than them cereal boxes.
Image 1 of 6
The site was pleased with some of its results for “black and white, filmstrip, pained face, cyclist, bike on bike”, suggesting they could look great on an indie rock album cover. Stability AI dream studio (opens in a new tab) was less successful at rendering the prompt “Exhausted cyclist slumped on a bike with sweat on his forehead with dark skies and lightning”. This poor guy is going to need a serious physio.
Image 1 of 2
So what was wrong? One thing that could cause problems is that most of the generated images are horizontal. Most AI image generators currently can only generate square images. The Stable Broadcast allows different aspect ratios to be chosen, but, due to the images it was trained on, it often doesn’t seem to do a very good job with images that are wider than they are tall. He often fills the extra space with duplicates, giving you two of whatever you asked for.
Some descriptions can also generate strange results. It seems that some generators take “close-up” to refer to the closeness between subjects in a composition rather than the type of shot. And the AI doesn’t know what feels fair or realistic. He doesn’t know how many wheels a bicycle should have, for example. Their training data undoubtedly included images in which the wheel of one bike was hidden behind the wheel of another or in which not all of a person’s fingers were visible.
Some of the glitches and artifacts in the AI-generated images can be fixed in editing software, but in this case many of them are complete, just like the mangled bikes depicted. For more on how the different AI image generators see this comparison of the best AI image generators.