ARKit 52 face shapes on a realistic guide
The 52 ARKit face shapes on an existing character, with the approved look kept, so the face works with iPhone capture, lip sync tools, Unity and Unreal.
- Role
- Face setup and shapes
- Tools
- Blender, Cycles
- Scope
- Test task plus repair passes
- Delivered
- .blend, FBX, preview
-
Shapes at full strength, named to the ARKit spec, including jawOpen and mouthFunnel. The last tile combines a smile, a half blink and raised inner brows.
-
Close-up of the eyes: blink closes onto one clean line; looking out and up stays clean after the eyeball re-texture.
-
Close-up of the mouth: jawOpen at half and full, funnel, pucker, a speech combination, smile, press and side shapes, with teeth, tongue and a dark mouth interior.
The brief: prepare the full ARKit set on an AI-generated guide character without changing her approved look, then calm, natural facial animation and lip sync. The client and project are not named.
What the model needed first
- The left eye was fused to the lids: both eyes became separate objects that rotate for gaze.
- Up to 2 mm of asymmetry: evened out by moving vertices, not rebuilding the mesh, so UVs and textures stayed untouched.
- Lips shared vertices along the closing line: split so the mouth can open and press.
- Teeth and tongue added as separate objects that follow the jaw.
Repair passes: a hole at the outer corner of the left eye closed with the original faces and UVs; the lip seam straightened and split into proper upper and lower lip edges; the mouth interior rebuilt (inner lip roll, dark mouth bag, teeth and tongue that follow the jaw); lid shapes rebuilt to slide around the eyeball; the eyeballs re-textured so a rotated eye stays clean. Nothing outside the eye and mouth zones moved; all 52 names, the face UVs and textures are unchanged.
Status: all 52 shapes work, including jawOpen, mouthFunnel and a speech-like combination, shown in the clip as a few syllables. Still to polish: a small fold at one mouth corner, the full-open mouth is slightly rectangular and mouthFunnel reads as a slot rather than a round “O”.
How I work: you send the model (.blend or FBX) and one voice line; I prepare the face and the shapes, then a short lip sync test on that line so you can judge the speech before the full set. Communication in writing.