How to Select Winning Patches

Once you’re satisfied with your prompt for AI, it’s time to generate patches. Since each batch of 4 patches only take 5~ seconds to generate, selecting the perfect patch can save you a lot of time down the road, eliminating the need to redo the entire PBR generation process multiple times with different patches.

Rather, it’s often more efficient to quickly generate 5-15+ batches of patches until you find the one.

The following examples provide some good tips on how to select winning patches.

Potentially Poor Patches

For materials like rooftops, tiles and bricks, you’re typically looking for even and aligned panels. In this case, the shingles are misconstrued in different sizes.

For finer, more patterned surfaces, the AI often has difficulty reproducing details, and can make the patterns look blurred and mashed together.

The grooves from this bark look unnaturally jagged. On first impressions, if you can’t imagine a patch looking diverse and natural in 2K, ignore it and move on.

This is an example of an extremely close-up surface with a distinguishable area of depth. Patches like these will typically look very repetitive when tiled.

Potentially Winning Patches

As far as AI generated textures go, the ridges are clean cut and even, the patch is zoomed in enough to capture the details of the shingles without being too repetitive.

Sharp and well defined patterns will make for better tiling and upfront quality overall.

From afar, the patch looks natural, with intricate depths and coloring that will result in fantastic PBR surfaces.

The patch is distant enough to not be overly repetitive, while maintaining diverse layers of detail that emphasize the material’s condition.