v0.1 · live in production · Frankfurt

Eyes that actually look back.

An avatar API where the iris is awake. Fifty-two ARKit-compatible blendshapes, per-eye gaze vectors, and micro-expressions — streamed in fifteen milliseconds. So your avatar doesn't just see. It watches.

Request access → Open the live demo ↗

15.0ms

per frame · CPU

52∙ shapes

ARKit-compatible blendshapes

478∙ pts

3D facial landmarks

1∙ rig

Unity · Unreal · Three · VRM

response payload

A frame is a face,
not a guess.

Send a JPEG, a base64 string, or a webcam frame. Get back the same schema every time: per-eye gaze, openness, squint, dilation hints, a full ARKit blendshape table, head pose, and a confidence stamp.

Drop the JSON straight into Unity's SkinnedMeshRenderer.blendShape, Unreal's MetaHuman ControlRig, or any VRM rig. Zero glue.

endpointPOST /v1/analyze

latency p5014.8 ms

throughput~68 frames / s / core

schemastable · semver

analyze · stream · /v1 LIVE

the moat

Other APIs detect a face.
We detect a glance.

01 / saccade

The flick that makes a gaze feel inhabited.

Real eyes never sit still. Eyeconic streams the micro-jumps — two to four per second — that separate a watching avatar from a sleeping puppet.

3–4 saccades / sec · sub-frame timing

02 / vergence

Two eyes. Independent. Properly converged.

Per-eye gaze vectors with sub-pixel iris localisation, so close-up vtuber shots and AR headset cuts never go cross-eyed.

left + right separately · vector + direction

03 / dilation

Pupils, lids, asymmetric blinks.

Dilation, squint, eye-wide, asymmetric blink — the signals that turn a polygon into a person. Twelve eye-specific blendshapes, mapped to ARKit names you already rig against.

12 eye shapes · 52 total · stable schema

and then · it lives on a face

The eye is the moat.
The avatar is yours.

Eyeconic doesn't ship a character. We ship the one piece every character needs. Drop the same payload onto a VTuber rig, a virtual influencer, a brand spokesperson, a game NPC — they all wake up.