Generating realistic landscape images with Nvidia Canvas showcase app

Nvidia Canvas is a showcase application that allows you to generate realistic landscape images with AI. Using options like grass, hill, ocean you can tell the AI to generate given feature on selected spot of the image. Lets take a look.

You can download Nvidia Canvas for free but to use it you need a PC with Nvidia RTX GPU.

The UI is quite simple although unique to this application:

Nvidia Canvas UI

Nvidia Canvas UI

You start from selecting an image from the right column. You can add your own as well. You should use landscape images, on others you will get weird abstract results. I used Windows XP wallpaper.

When you select an image you likely won't see it on the preview. The app has two panels - drawing box on the left and preview box on the right. Using materials from the right column you draw where given landscape feature should be on the image and the AI will try to draw it there.

To have a base start you should try drawing where original landscape features are on the base image - so I've market the area where the Windows XP wallpaper had grass and where it had sky. This generated something similar to the original image. Next I started experimenting with adding water, ocean, hills and so on. Note that you need somewhat realistic angles and size of features like hills or mountains for the AI to generate something that looks like them and not weird artifacts.

So here are some of the examples:

Water and sea added

Water and sea added

Some hills in the background

Some hills in the background

Or mountains

Or mountains

As you can see using basic painting methods we can generate realistic images of landscapes. Nvidia more advanced models demoed during various presentations can do similar feats with other type of images as well. This opens a lot of options for image and video processing and making such tasks much easier for less advanced users. Already NPUs in mobile phone chips allow using trained models to augment such tasks.

RkBlog
Comment article
Comment article RkBlog main page Search RSS Contact