https://blogs.nvidia.com/blog/2021/06/23/studio-canvas-app/

Turning doodles into stunning landscapes — there’s an app for that.

• Why do you consider this synthetic media?
• What can you find about how this was made (if you can)? If it’s machine learning based can you find information about the data set used to train it?
• What are the ethical ramifications of this specific example (if any)?

Being able to turn a simple doodle into a stunning realistic portrait with an AI art application is a great way for people to rapidly and iteratively make changes to a scene to make it fit their vision. None of these scenes are actually real, they are all completely generated on the fly by the AI.

After doing some digging we can see that the dataset that the Studio Canvas app is trained with the COCO Stuff Dataset - The final dataset included 164,000 images that has 172 classes.

Welcome to official homepage of the COCO-Stuff [1] dataset. COCO-Stuff augments all 164K images of the popular COCO [2] dataset with pixel-level stuff annotations. These annotations can be used for scene understanding tasks like semantic segmentation, object detection and image captioning.

One example that pops to mind would be say a person draws a single line and it creates a whole image and then that person copyrights it. How would that work if another person draws a single line and now the first person says it's their creation. Also does it belong to that person itself or does it belong to the person that took the original image. It would be interesting to think about if the scene was generated each time to be unique so that the chance of copyright could never happen.