Enterprise Face & Person AI
Bring face and person AI to your products, with commercial models and APIs that run on your infrastructure.
Trusted by teams building the next generation of face AI
Solutions
Build with InsightFace: face recognition, person analysis with body ReID, self-hosted APIs, video face privacy, face swapping, and commercial model licensing.
Commercial licensing for InsightFace open-source model packages (buffalo_l, antelopev2, buffalo_s, buffalo_m, etc.) and proprietary high-accuracy models. From research to production deployment with full legal compliance.
Optionally enable RGB liveness checks before face recognition through the Python package or InsightFace Server.
Detect people, match registered identities through faces, and use body ReID when a face match is unavailable. Run locally with Python and Cheetah models.
Keep video face redaction on your own device. Track faces, choose who stays visible, and export blur or mosaic with the desktop app, CLI, or Python.
Commercial model license required
Case Studies
From identity verification and access control to creative AI and edge deployment, InsightFace helps teams evaluate, license, and deploy face AI models responsibly.
A fintech team evaluating face recognition models for identity verification, threshold tuning, and production deployment.
An access control provider exploring lightweight face analysis SDK deployment across embedded and edge devices.
A consumer application team reviewing face swap API access, licensing requirements, and responsible use controls.
Anonymized scenarios derived from typical enterprise engagements. Specific customer names are not disclosed for confidentiality reasons.
Licensing & Pricing
From open-source research to model licenses, local live identity deployment, production APIs, and SDK access — choose the licensing option that fits your business needs.
Custom
Commercial licensing for face recognition, person analysis, and face swapping models
Integration
Whether you prefer Python, C++, or REST — InsightFace integrates into your stack with minimal friction.
Use FaceAnalysis for face workflows, or PersonAnalysis to combine person detection, face matching, and body ReID.
import cv2
from insightface.app import FaceAnalysis
app = FaceAnalysis(name="buffalo_l")
app.prepare(ctx_id=0)
a = app.get(cv2.imread("a.jpg"))[0]
b = app.get(cv2.imread("b.jpg"))[0]
score = a.normed_embedding @ b.normed_embedding
print("Cosine similarity:", round(score, 4))
import cv2
from insightface.app import PersonAnalysis
with PersonAnalysis(name="cheetah_s") as app:
app.register(
"alex", "face.jpg",
body_images="body_crop.jpg")
frame = cv2.imread("scene.jpg")
for person in app.match(app.get(frame)):
print(person.person_id, person.matched_by)
import cv2
from insightface.app import FaceAnalysis
from insightface.model_zoo import get_model
app = FaceAnalysis(name="buffalo_l")
app.prepare(ctx_id=-1)
source = app.get(cv2.imread("source.jpg"))[0]
target = cv2.imread("target.jpg")
swapper = get_model("inswapper_128.onnx")
cv2.imwrite("swapped.jpg", swapper.get(
target, app.get(target)[0], source, paste_back=True))
Compare two single-face images with cosine similarity.
Native SDK for mobile and embedded devices. One API across all platforms.
// Initialize the SDK
HResult ret = HFLaunchInspireFace(path);
// Create a session
HFSession session;
HFCreateInspireFaceSession(&session, mode);
// Detect and recognize
HFDetectFaces(session, image, &faces);
HFExtractFeature(session, image, face, &feature);Self-host detection, optional RGB liveness, comparison, enrollment, and exact person search on CPU or NVIDIA GPU.
# Authenticated deployment; omit -H when auth is disabled.
curl -sS http://localhost:18097/v1/detect \
-H "Authorization: Bearer ${INSIGHTFACE_API_KEY}" \
-F "image=@photo.jpg"Research & Publications
Published at top-tier academic venues. Every algorithm in InsightFace is peer-reviewed and reproducible.
CVPR 2019
The foundational loss function for face recognition, setting new benchmarks on all major datasets.
CVPR 2020
Practical single-stage face detector with joint face detection and landmark localization.
ICLR 2022
Efficient high-accuracy face detection with NAS-based architecture search.
CVPR 2022
Scalable training strategy enabling face recognition with massive identity pools.
ECCV 2020
Handling noisy training data by learning multiple sub-centers per class.
Blog
Stay up to date with the latest developments in face AI technology, research, and applications.
Three papers address the production gaps around an identity edit: matching scene light without repainting skin, making diffusion editors actually use the reference, and cutting visual dubbing to two denoising steps.
Read moreThree September papers examine where face recognition breaks outside standard leaderboards: demographic score density, visible-to-thermal matching, and nearly indistinguishable identities.
Read moreSeptember's detection stack moves from boxes to reliable geometry: two landmark systems target mixed, difficult datasets, while a classroom study measures the end-to-end detector and recognizer trade-off.
Read moreWhether you need face recognition model licensing, face swapping commercial rights, a live streaming identity deployment, a local real-time identity workflow, InspireFace SDK, or custom AI cooperation — contact us to start the conversation.
We review commercial, enterprise, and research cooperation requests. Please use a company email address when possible.