AI 3D Character Creation Pipeline

Step 2 : 3D Model Generation

3D model generation is the stage where AI systems convert approved character
concepts into a three-dimensional form using image-to-3D or text-to-3D models. Based
on reference images or detailed textual descriptions, the AI generates a base mesh that
captures the character’s overall shape, proportions, and primary forms. This base mesh
then undergoes automatic topology optimization, where the geometry is reorganized
into cleaner edge flow and evenly distributed polygons, making the model suitable for
further refinement, animation, and integration into real-time engines or rendering
pipelines.

For model creation we will use Tencent 3d

Its a chinese platform, make sure you translate the page when using

We need to use the 3d studio over here as it generate the best results in comparison to
the default AI module and other 3D generation platforms. Initially you will have to
register yourself first to get the access, which normally is granted after 1-2 days (notified
via registered email)

Step 2.1: Generating Mesh based on the Reference

Once access is granted you can start uploading your images in the “Geometric
generation” tab and generate a high to mid poly mesh

Once the mid poly mesh is generated you can review for details from every angle.
Sometimes, the AI algorithm might cause issues in the generated mesh. This could be
the case if the reference image has any artifacts or its of lower resolution or quality.
Make sure to use high quality and clear images for generation of meshes

Step 2.2 Low Poly Mesh Generation or Retopology

This AI platform will allow you to create a low poly version or a retopologized version of
the high poly mesh. Depending on your game optimization budget, this is the stage
where you will determine how many polygons you game model will be using. Once on
this stage generate a low Poly Retopologized mesh.

A low poly version of the previously generated hi poly mesh of the same character has
been generated using the Low-Mode topology tab on the Tencent 3D Studio

Choose the options Low or Middle and check the Four sides option for a Quad mesh as

the character is supposed to be rigged and animated

Step 2.3 Automatic UV Generation using AI

UVs are very critical as they are used for texture generation and painting. Other AI
platforms though provide this functionality but the output they produce is unadvisable to
be used in game. Tencent on the other hand does a pretty reasonable job in doing the
UVs. We need to make sure to keep the UV island as minimum as possible keeping the
game performance and optimization in mind. The more the UV islands, the more draw
calls.
Please note that if you are not satisfied with the automatic uv generation (although it’s
pretty good to go) make sure you do the UVs manually using your preferred 3d
software.

UVs shown on the left screen, generated by Tencent’s AI algorithm

Please note that the model on the left may seem that it has been triangulated, but when
you will download the mesh and open it into your preferred 3d package, the mesh will

be of Quad topology.

Step 3 Texture Generation using AI

For texturing we can use the reference image or use a prompt for the platform to create
textures on the uvd mesh for us. Though tencent AI does a pretty reasonable job and it
can be used. Either u can use the generated model or upload an already created model
from your local drive. Below is an example of the texture generated of the same bear
warrior character giving the reference image.

Use the Tusheng texture option for uploading a texture reference image and generating

a texture based of that image.

As an alternative to Tencent we can use Tripo AI for texturing as well. It does a good job
for texturing, but for modeling and character generation I would not recommend it.

Tripo AI also does a resonably good job. You can generate PBR textures

Once texturing is completed, you can download the model along with the textures.

Step 4 : QA of the model and texture

Once the above steps are complete, it is time to review the model and do the necessary
changes if required. Those changes or fixes could be in the mesh topology, which if not
fixed can lead to issues in rigging and then animation and the second in the texture.
Sometimes the textures are not generated properly or they have artifacts which can
require some tweaking from any 2d paint program like photoshop or a 3d painting
program like Substance painter.

Tencent does a reasonable job in the mesh topology quality but a close inspection is
always a welcome as it can save you time later in the production process.
Look for inconsistencies and missing loops that are required in the deformation areas,
like neck, legs , arms and fingers. Add additional loops if not there already.

Also if required fix the issues in UVs, that might include overlapping or multi island
separated UVs. if possible fix them or stitch them to minimize the UV island distribution.
In this case you will have to reupload the mesh on the AI platform and regenerate the
texture over there. The new texture will be according to the UVs you modified and fixed.
If everything is OK, its time to make sure that our model is ready scaled properly before
using it for rigging and animation.

You can scale the character to world units using feet or equivalent units in cm. For this
character it is roughly scaled at 6 feet or 182.88 cm
Once scale is fixed make sure the characters transform stats are set to zero. In other
words Freeze transformations (Maya) , Reset XForm (Max) and Apply all transforms
(Blender).
Once all of this is done, you can rig your model and animate it. And Use it in any game
engine you like.

Bear character model generated with AI and textured with AI ready for game use