If you’ve ever built a character from reference photos, you know the part that eats the day isn’t sculpting the likeness, it’s the mesh housekeeping afterward. KeenTools, which has made the FaceBuilder add-on for Nuke and Blender for years, now has a standalone service aimed squarely at that problem. KeenFace was covered by Digital Production on October 22 and in CG Channel’s roundup on October 25; neither source gives an exact launch date.
What changed
KeenFace reconstructs a textured 3D head from photographs, either in a browser-based Studio or through a cloud REST API. According to KeenTools’ documentation, as reported by Digital Production, it takes between 1 and 15 images. The capture guidance asks for a front view plus turned views from each side, and a single photo is allowed when that’s all the reference you have.
What comes back is built for a pipeline rather than a one-off render:
- Consistent topology. Every head uses MetaHuman topology, and the maker says vertex order and UVs stay the same from session to session, so heads are interchangeable in a rig or a blendshape setup.
- Textures. 4K PBR maps (base color, normal, roughness, specular), with an option to clean out hair and fill gaps, or keep the raw photographed look.
- Expressions. 51 ARKit-compatible blendshapes adapted to each head.
- Formats. The Studio exports GLB, OBJ, PLY, STL, and FBX, and there’s a MetaHuman export meant to go through MetaHuman Creator’s Conform from Template.
Pricing is per head rather than a subscription: see the sidebar.
Why it matters for designers
For 3D generalists, game and AR teams, and anyone making digital doubles, consistent topology is the real headline. A head that always arrives with the same vertex order and UVs can drop into an existing rig, texture workflow, or set of expressions without retopology. That’s the step that usually turns a quick photo scan into a multi-day job.
It’s also a sign of where AI-assisted 3D is heading: less “generate something that looks like a head” and more “give me a clean, standard asset I can keep working on.”
What to watch out for
Almost every specific claim here comes from KeenTools itself. Digital Production says plainly that it hadn’t timed or tested the service, including the claim that reconstruction takes minutes and that topology stays consistent across heads. Check likeness, how the face deforms, and how the materials read against your own assets before committing.
The maker’s own capture guidance lists a lot that can trip it up: hats, hoods, glasses, long hair, hidden ears, deep shadows, blown highlights, and a subject who changes expression between shots. Even, indirect light and a still, neutral subject matter.
There are gaps in the exports, too. The MetaHuman package doesn’t include eyes, teeth, or blendshapes, the ARKit set needs GLB and lacks tongueOut, and hair grooming is still a planned feature, so you get a bald base head. Heads made in the Studio and through the API share credits but can’t be opened in the other one.
Finally, this is a real person’s face going to a cloud service. Get consent, and read the terms before you upload anyone’s photos.