SkinTokens in ComfyUI: rig characters for Unreal Engine 5

This tutorial follows SkinTokens in ComfyUI from installation to a playable Unreal Engine 5 character. The presenter describes the rigging tool as open source under the MIT license and says it runs locally with 6GB VRAM. The supplied requirements specify an NVIDIA GPU. The Windows setup uses ComfyUI Easy Install, PyTorch 2.8, CUDA 12.8 and Python 3.12, with Blender 4.2 or later installed and available on PATH.

The rigging examples include a wolf-like humanoid, a dragon and a dog. The presenter reports a 180-second run for the dragon, whose rig includes wings, four legs and a tail. Automatic skin weights still need inspection: the first character's tail moves with its leg, and the tutorial corrects that in Blender's weight paint mode.

A longer asset workflow uses Trellis 2 for generation, Hunyuan 3D Studio for retopology and UVs, texture projection, and Modddif for texture cleanup before returning to SkinTokens. Those website steps mean the complete demonstrated workflow extends beyond local AI processing. A sponsored Customuse segment also presents a concept-to-asset workflow.

For Unreal Engine 5, the presenter selects the matching export mode and recommends GLB because FBX causes a scaling issue in this example. The final steps create a Control Rig, retarget animations, fix squashed legs in the animation blueprint, and adjust the character's scale and location for third-person play.