WaldoHunt

• 

• 

• 

• 

• 

• 

  T

Inspiration

Was always inpired by the Where's Waldo images and the Hidden Object games of old

What it is An AI generated stylized and photorealistic images of crowded scenes are generated and hidden "objects" and persons within the clutter are hidden. There is a time count down to find our hidden "Waldo" before the image generation process triggers a new scene. Ultimately, there is an overarching plotline still eked out. But the user needs to be able to solve a series of puzzles from a random series of AI Generated images within a given time.

How we built it

The underlying backend is still being built since creating an AI Generated Image that consistently hides a target object is a technically daunting task.

Challenges we ran into

Like describe above, it is easier using text to image prompts to create an AI image. Image to Image generation with where a we also needed to include an in-painting feature for our "Waldo Object" and also be able to "mask" or hide it sufficiently to make the game engaging enough was quite the task,

Accomplishments that we're proud of

We have learned a lot in the process, especially regarding the testing of various local AI image Generating models with less powerful machines for our given task. But the challenge has been to migrate the process to some known cloud based AI service.

What we learned

Proving a concept on a local machine is not always the same as replicating the same on the cloud. And consistent and stylized AI image gen can be hard.

What's next for WaldoHunt

We are still re-working the AI image gen parts of it to make the whole simulation more engaging--especially where a group of people can challenge themsleves to be the first to find the given hidden object in a virtual casual social context.

Built With

• amazon-web-services
• bedrock
• diffusion
• gemini
• llama3
• stable