I created an AI Art based Wordle type game calledDiffudle (Dead, unfortunately). In short, you are given images created using hidden “prompts” which you have to guess. For the MVP, I used to generate images myself and upload them to a database. Between selecting prompts, styles, and images; it takes me around 30 mins to do this. Since this is a daily game, doing this every day quickly became a timesink. I realized this is a crisis which I need to solve.
I used to visit generative art sites likelexica.aito search for similar prompts. I used styles from other prompts to generate images. Eg: If the prompt is “cavemen selfie”. The actual images will be generated from the following phrases:
Here's the actual images generated for the above game. The images are stylistically different but all of them are still kind of doing the same thing.
I started using ChatGPT to create prompts with immediate results. I asked GPT to generate phrases for me with the given prompt and here's what it came up with:
Impressive!! It contains different aspects of the same prompt which can be used to generate images. Here's the final images. You can see that the images now describe the prompt instead of just drawing it.
At this point, I could've automated the whole game:
Since all this is paid, I decided to wait until this project starts generating some revenue before buying a subscription to these.
Thanks for Reading!! Reach out to me prakhar897@gmail.com for any queries.