Know your memory limits (BTS #22) (Patreon)

Pre-rendered images involve some unusual skills during the process of making them. Video game engines are relatively efficient in that they are able to automatically hide objects on the fly and scale the quality as needed. That is because they are designed to operate on a real-time basis where visual fidelity is not the highest priority. Those requirements do not apply to Daz, and VRAM management is a skill I've had to learn in order to accommodate my RTX 2070s limitations.

Character models, hair, and textures are the main things that eat up memory. Obviously, you want as many parts of a character to be as high quality as you can, but there are many ways to keep the quality while saving a significant amount of memory. I'll explain using a scene from Ménage à Moi where four characters are sitting and talking at a table.

The bookshelves, tables, and chairs are mostly instances and the room itself is a relatively simple structure. Neither takes up much VRAM on their own. On the other hand, four characters required a bit of manual optimization on my part. Each human model is split into the mesh (the structure itself like the face shape) and the mats (AKA textures for the skin, eyes, etc). The same applies to each piece of clothing, jewelry, and accessories.

In this scene, there is one part of these models that are consistently hidden from view. The legs. While I designed these outfits to include footwear, they weren't visible so I deleted them. Next to go were the mats for the legs as each section of the skin on a G8 model is a set of 4k images. Spread over four models, this is a substantial memory saver.

Daz renders everything in a scene, whether a camera is looking at it or not. While strange looking, by hiding things (in this case, parts of the body) I am telling Daz to ignore it while rendering. Additionally, certain other common problems like poke-through on clothes are lessened or eliminated.

For those of us with moderate hardware, making scenes is not a simple matter of plopping down the characters, posing them, and then moving the camera. As you can see, there are a whole lot of considerations that need to be taken into account. If you ever wondered why some VNs used solid color (usually black or blue) models as background extras, memory limits are one of the primary reasons.