Today, I decided to embark on a little experiment with artificial intelligence, diving into the realm of video game graphics. I was particularly intrigued by the challenge of recreating the famous victory screen from the game PlayerUnknown's Battlegrounds (PUBG) — the "WINNER WINNER CHICKEN DINNER!".
Using Chat GPT DALL-E I tried to generate a very similar image trying to resemble the screenshot I initially sent. Here is the initial result
As you can see it is very obviously a reference to the game PUBG but a lot of the GUI/UI aspects are clearly in the wrong place. My expectations
were that the graphics were going to be obviously AI generated but the GUI would mostly stay in the same area and resemble the original picture
I sent.
I tried addressing those issues that I just listed and I again prompted DALL-E to try with the corrections.
None of the corrections I mentioned were shown on the image DALL-E generated. It was actually worse than the first one because the classic phrase "winner winner chicken dinner" was not included. Instead, it says "chicken giinner!". Now with this obvious failure, I went to ask Chat-GPT to write a blog about what it thought about how successful it was in generating images that were similar to the original.
What I really found interesting was that it interpreted what I considered a failure to adhere to the prompt as it was taking a creative take on gaps left for interpretation. Hence why it was different in many aspects to the original screenshot I initially sent.