auraface-v1

by isidentical

Open-source face recognition with ArcFace architecture — commercially licensed

180Kdl/month

65Mparams

HuggingFace Use in Pipeline

Identifiers

Model ID

isidentical/auraface-v1

Feature URI

mixpeek://face_identity@v1/isidentical_auraface_v1

Overview

AuraFace v1 is an open-source face recognition model built on the ResNet100 architecture with Additive Angular Margin Loss (ArcFace). Developed by Batuhan at fal.ai, it extracts 512-dimensional face embeddings that map facial identity into a compact vector space where distances directly correspond to identity similarity. Unlike the original ArcFace model which has non-commercial restrictions, AuraFace is released under Apache 2.0 for unrestricted commercial use.

The model achieves 99.65% accuracy on the LFW benchmark and provides a strong balance between recognition accuracy and commercial usability. On Mixpeek, AuraFace powers face-based identity search across video and image libraries — find all appearances of a specific person, cluster faces by identity, or verify identity matches across content collections.

Architecture

ResNet100 backbone with Additive Angular Margin Loss (ArcFace). Trained on commercially and publicly available face data. Produces 512-dimensional face embeddings. Compatible with InsightFace framework for CUDA and CPU inference.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "media-library",
  source: { url: "https://example.com/interview.mp4" },
  feature_extractors: [{
    feature: "face_detection",
    model: "isidentical/auraface-v1"
  }]
});