AIRI and Blender: build a talking 3D AI companion

This tutorial follows an anime character through mesh cleanup, texturing and rigging into a talking desktop companion in AIRI. The speaker generates separate body parts to make editing easier, then assembles them in Blender. Geometry changes come before UV unwrapping because they can disrupt the texture mapping. When low-poly hair generation fails, he sculpts a higher-detail version and retopologizes it.

The avatar uses a Mixamo rig, with manual corrections to bone positions and weights. The Blender VRM add-on handles bone mapping, spring bones and colliders for the braids and cape. Faceit, a Blender add-on, creates facial shape keys that also need VRM expression bindings. MToon materials provide toon shading and adjustable outlines. The speaker recommends checking the target application's VRM requirements; he says version 0 works better for his VSeeFace setup, while other uses may require version 1.

Project AIRI is presented as open source, with web and Windows versions and support for 2D or 3D characters. Its provider settings allow local models or external APIs. The demonstration uses connected APIs, so it does not establish an entirely offline workflow. The speaker uses Claude to customize his fork with extra animations and a separate animated GLB object. His desktop version has a transparent background, resizing and hiding when the pointer covers the avatar; those details belong to the demonstrated fork.