CompreFace Review

CompreFace is a ready-made face recognition service from Exadel. It runs as five Docker containers (core, API, admin, UI and a Postgres database) and exposes a REST API for face recognition (1:N against subjects you upload), face verification (1:1), face detection, and plugins for landmarks, age, gender, mask detection and head pose. A web UI handles applications, services, API keys and user roles, so you can give different teams access to different face collections. SDKs exist for JavaScript, Python and .NET, and it is a common add-on in Home Assistant setups via Double Take and Frigate.

Setup is: install Docker and Docker Compose, download the release archive, run docker-compose up -d, and open port 8000. A single-container image and Kubernetes scripts are also available. The default build uses FaceNet with MTCNN detection; custom builds swap in InsightFace models (MobileFaceNet, SubCenter-ArcFace) with optional GPU support. All data stays on your machine in a Docker volume.

CompreFace is free and open source under the Apache 2.0 licence, with no paid tier listed in the repository.

What it costs to run: your own hardware. The default build needs an x86 CPU with AVX and does not support GPU; the InsightFace-based builds need AVX2, and the GPU variants need CUDA. The docs advise GPU builds for real-time recognition and mobile-derived models for older hardware. The project publishes no RAM or throughput figures.

If you use an InsightFace-based build commercially, note that InsightFace states its pretrained models are for non-commercial research only unless licensed.

Prices checked on the vendor's pricing page on November 12, 2026. See the CompreFace repository and licence.

From the CompreFace README and docs:

CompreFace suits teams and hobbyists who want a working face recognition server on their own hardware without writing ML code: door cameras with Home Assistant, attendance or access systems, internal photo tools. Role management makes it workable for small organisations.

The catch is maintenance. The last release (v1.2.0) is from August 22, 2023, the last commit on the main branch is from November 2023, and there are about 235 open issues, so plan to own security updates yourself. For an actively developed library, use DeepFace; for an actively developed server with a commercial model licence option, InsightFace Server; for a managed API, Amazon Rekognition.

CompreFace has about 8,400 stars on GitHub. Home Assistant users recommend it with Double Take for doorbell and camera recognition; one user on the Home Assistant forum says they are satisfied with the results but gave its VM 6 GB of RAM. Tutorials such as this Frigate, Double Take and CompreFace guide show it as the recognition step in self-hosted camera setups.