AI-powered image generation models are advancing at a rapid rate, but it’s still common for them to churn out questionable images. Since it’s easy to assume that the human prompts are the problem, I decided to test if AI has an easier time working exclusively with AI-generated prompts.
Rules of the Experiment
When AI image generation models hit the scene a few years ago, we all thought it would toll a bell for all the people working with visual media. This didn’t end up being the case, though. Despite being able to generate hyper-realistic photos, AI images would often fall into the category of unpredictable, especially if you required something a bit more complex (AI tends to struggle with hands, for example).
You can either blame AI models themselves for this issue or the fallacy of humans and our inconsistent prompting skills. The natural way to test who’s to blame is to see if image generation models deliver better results if you input generated prompts.
