In this art experiment, I compared the illustrative abilities of AI with my own. I illustrated a location from one of my dreams and then described it to AI to see how the results would differ. I chose this subject because the image has been vivid in my mind since I had the dream in elementary school and because it is not a real location. If I used a real location, AI art generators could use internet references in their designs, and that would change the results. I created my version of the scene first,  so it is untouched by how AI might generate it, and the result is accurate. Then I used the AI art generators to create their versions of a world that they have never seen. 

​

Both of the generators struggled with the shape of the platform, and once I got the shape in OpenAI using DALL·E, DALL·E  did not want to rotate it on its side. The videos show the different ways I tried to describe the dream location to the AI generators and how they misinterpreted the prompt. It was easy for them to generate a “one room house with a red tile roof and a wooden front door,” but the AI art generators were not able to effectively combine this image with the specific shape of the platform. I have chosen the best version from each AI art generator to be displayed.