Lab Rats 2: Even More Automation!

If you've been following me here on Patreon for the last few months you'll know by now that I have a deep love of automation. I've spent a couple of my prototyping weeks in the past creating the automation that let me get Lab Rats 2 to it's current stage, and now that it's development has stepped up to full time I've had more time to improve things even further.

First, I should explain how character images for Lab Rats 2 were generated up until today. There are three base body types: average, thin, and curvy. Each of these had a corresponding model in Daz. There are also ten breast sizes ranging from AA to FF. These also had a model in Daz, and had the same proportions as the average body type. For each upper body piece of clothing a render is made with all 13 models. The 10 breast images are edited down so that they are only the breast section, while the 3 body images were edited so they would layer under the breast images properly. If everything worked out the resulting piece of clothing looked natural, while limiting the amount of render work that needed to be done and keeping the total number of unique images to a minimum.

That design worked well, but it had a few problems: It was hard to capture self shadowing from large breasts, so clothing seemed to cling to a girls chest more than it really did. Different postures, especially ones from the side, would make layering very tricky to work out. The biggest problem though was the amount of time I had to put in to edit down each breast image so that it would layer correctly. While it was relatively simple work there was no way to automate it so that I could focus on other tasks.

The solution to all this happens to be quite an easy one, and it's a favourite of programmers everywhere: brute force! Each piece of torso clothing (the only one that needs to account for breast size) is now rendered 30 times, one for each combination of chest size and body type. This removes the need for image editing entirely (beyond cropping the initial render) and lets me automate the entire process from start to finish. The trade off is a 2.3x increase in render times and image size for tops. The increased render times shouldn't be a problem, and the total image size per top will still be less than a single 1920x1080 background image.

Now that the image generation process is automated I'm free to make changes to the current models I'm using without worrying about wasting hours redoing editing work. As a proof of concept I've reworked the thin and curvy body types to be more pronounced than they were before and re-rendered the entire library of clothing I've got so far (a total of 22 items). Using low quality render settings the entire process took just over two hours, and most importantly didn't require any input from me. Now that I know everything is set up right I'll be producing high quality renders (which will take more like 15 hours) for you guys to take a look at some time in the next few days.