For a character that is “user facing” I wanted to make an over the top female robot. Someone that would have replaced waiters, clerks and all service professions in general.
Prompting
This character turnaround was created along the other characters when I experimented with getting the generative AI to produce consistent images of the same character from multiple angles. In the prompt text I even have a typo, but that did something so it was allowed to stay in. I chose an image with no back view mainly because I needed a delicate robot like this and the AI just would not draw one with a back view. Also it again was a good test to do, to see how a missing back view would affect the work.
3D modeling
This time the added challenge for the modeling and eventual texture projection came from the fact that there was no back view. I painted the missing surfaces by hand in photoshop.
The modeling process was not difficult. Following the concept was easy because of the direct views. The turnaround images did not quite match, but I always chose to add the details I liked the best. The character had quite a bit of pretty vague details and a lot of them were left to my interpretation. Often times I just modeled pretty freely without worrying too much about the concept. I likes the contrast of smooth, round shapes and the sharp corners. Some time was spent on making the round shapes extra round and the sharp shapes extra sharp.
I only modelled half of the character and mirrored the result. Saving some time.
I captured the whole process from start to finish. But after reviewing the footage I found it is VERY boring to look at. Just a zoomed in view of weird shapes with camera flicking from side to side like maniac. But here it is for your viewing pleasure: 6 and a half hours of modeling and photoshopping condensed to one minute!
This is how the finished model turned out:
UV textures
The same trick was performed to get the textures baked on to the UVs as always: I created morph maps for the mesh that deformed the character to fit the AI generated images perfectly and rendered the image’s base color to the UV map. The UV unwrapping for this model was again really tedious as the model has many overlapping shapes and a fair bit of details. I also kept the model as one continuous shape for Mixamo auto rigger to work as well as possible.
Projected textures:
I combined these projections again to one map by using the best bits of both on different surfaces. As the back view was completely missing this one required quite a bit of overpainting and the whole back of the character was missing.
I also derived an emission map, a roughness map and a normal map from the diffuse channel data. I create these maps to break up the light on the otherwise low polygon surface. To add an illusion of details.
Final UV maps:
Rigging
I am fairly new to Mixamo rigging, but it is very simple and seems to be quite acceptable! This tool saves so much time with somewhat standard humanoid characters! Mixamo is also very good for previewing how the character will look with different animations. When I rigged the previous character, I used a mixamo rig I copied from another character as a base. As the rig works fairly well in Unity.
The end result
I am fairly happy with the end result. The model again is very low detail and the texture close up looks very low resolution. But viewed in-game that does not really show.
I have also never before modelled anything quite like this. It is a bit too sexualised for my taste, but the role in the game it is meant for requires her to be.
In this post I describe how to add a video texture to the cyborg’s visor.
Leave a Reply