#5 - Paintovers & AI Inspiration

My next step was to visualise the concept I had chosen in the context of my existing landscape.

For the sake of getting a better base to paint over, I have used an existing asset pack for the landscape and trees, which I intend to replace with bespoke assets going forward.

I used software called PureRef, and took advantage of it's "Overlay Mode" to display a cut-out version of the building from the concept on top of my UE5 viewport:

Doing it this way, instead of editing a screenshot using photoshop, meant that I could keep moving the camera around to different angles and positions, and adjust the overlaid image until I was happy with the composition.

I then took these screenshots into photoshop to paint over to get a better idea of the layout and details of the scene.

I am still very undecided on which of these two compositions to go for, as there are definitely elements of both that I prefer.

I like the river that flows through the foreground of the first shot, and the bridge works well for leading the eye, but I also prefer the silhouette of the building and weathervane against the sky in the second shot, and the long view of the valley running into the background I feel emphasises the scale of the world better.

Something that occurred to me during this process, is just how ludicrously huge the tower in the back is, relative to the world around it. I definitely need to do something different with that in the future.

As an experiment, I tried running my screenshots through Stable Diffusion, an AI tool that can use a text prompt along with an existing image to generate variations:

 I was impressed by how well it was able to figure out the general idea of the scene. Some of the shapes, colours, and smaller details it has brought in have really helped to provide ideas for how I can push the scene further.

Going into next week, I will be looking to finally get the building blocked out in 3D. I will be using my concept breakdown from the previous post to create a simple kit in Blender, which I will then try to recreate the concept with.