Over at one of my favorite Retro Game Forums stayforever.de, there’s a thread about turning retro game art and scenes into realistic photos, leveraging generative AI like ChatGPT or Copilot and the like.
What a fantastic experiment that is – I tried it out myself and enjoyed a good two hours, converting a number of screenshots from my repository into more realistic photos.
Here’s a scene from Loom that you might know. The original:
What ChatGPT made from it:
What Copilot made from it:
The prompt that I used is the same in ChatGPT and Copilot, the original screenshot is the same too. I found the lighting and the scenery done nicer in Copilot, I must say.
The prompt I used in both cases was this:
I have a picture from a retro game that is in pixel art. I need you to convert this picture into a modern, realistic image. Make it so that it could be a real picture that I am looking at. Make it realistic. I want you to preserve the colour palettes and lighting as much as possible, and I want you to keep as many details as possible. You can resize the picture and use a larger, modern resolution for the end result.It doesn’t always work out great, because clearly, some pixel art will lead the AI to misinterpret what’s in the picture. For example, in another scene, Bobbin was turned into an Astronaut:
FRICKELSOFT 