Best Face Recognition Search for Media Libraries (2026)
12 min
red
For a professional video library, the best face recognition video search sits inside a media platform your editors already use: Shade (our product), iconik on its Professional and Enterprise plans, Axle AI and Moments Lab all let you name people and then search for them. Amazon Rekognition, Azure AI Video Indexer and TwelveLabs are developer services, and two of them come with a catch: Azure requires Microsoft's approval for identification, and TwelveLabs' people search is in beta. Immich is the free, self-hosted option.
Face data is regulated as biometric data in several US states and in the EU and UK, so this guide also covers consent and the law. We checked every price on the vendor's own site in November 2026. We make Shade, so it is listed first.
Detection, clustering and recognition are not the same thing
Face recognition search lets you find every clip a person appears in. The software detects faces in your video, groups similar faces together and lets you attach a name to a group. After that, searching "Maria" returns every asset with Maria in it, often with the timecode.
Vendors use three terms loosely, and the difference decides what you can actually search for:
Face detection finds that a face is in the frame and where. It does not say who it is. Google's Video Intelligence API, for example, detects faces but does not name them, and Google is shutting that API down on September 14, 2027.
Face clustering groups faces that look like the same person, without names.
Face recognition (identification) links a face to a named person you or the system has defined.
Consumer and editing apps do versions of this. Google Photos builds "face groups" you can label. DaVinci Resolve Studio uses facial recognition to sort clips into bins by the people in the shot, which helps inside one project but is not a shared library. This guide is about professional video libraries: sports archives, news and documentary footage, agencies with talent and brand content, and production companies tracking cast and interviewees.
Six things to check before you buy
Video, not only photos. Many photo tools skip video or only sample frames. Daminion's face recognition, for example, works on images only.
Detection, clustering or named identification. Match this to what you need to find.
Where face data is stored, and whether you can delete it. Face vectors can count as biometric data under the laws covered below.
Approval and consent. Some vendors require approval before you can use identification at all.
Access from your normal workflow. A match is only useful if editors can reach it from the library or their NLE.
The pricing model. Per minute, per seat, per TB or quote-based change the math completely.
Eight tools compared
Tool | Type | Names people? | Works on video? | Pricing model |
Shade | Cloud NAS with AI | Yes, you name people and search by name | Yes (images and video) | Growth $29.75 per seat per month billed annually ($35 monthly); Enterprise quoted |
Amazon Rekognition | API | Yes, via face collections | Yes (stored video) | $0.10 per minute for stored-video face search; $0.00001 per stored face per month |
Azure AI Video Indexer | API and portal | Yes, but Limited Access approval required | Yes | $0.09 per input minute (Standard video preset, includes facial recognition) |
TwelveLabs | API and playground | Yes, via Entity Search (beta) with reference images | Yes | Free, Plus $29 and Pro $99 per month; Enterprise custom |
iconik | Cloud MAM | On Professional and Enterprise | Yes | Starter pay as you go; Professional and Enterprise quoted |
Axle AI | MAM and tagging engine | "Trainable faces" | Yes | Tags from $200 per month; Cloud $20 per TB per month; MAM from $2,995 |
Moments Lab | Cloud media platform | Yes, characters and talent you train it on | Yes | Quote-based |
Immich | Open-source, self-hosted | Yes, you name groups | Photos and videos (detection uses a preview image) | Free (AGPL v3 license); you host it |
The AWS and Azure rates are for US regions. For iconik, facial recognition is not on the pay-as-you-go Starter plan, so face search starts at a quoted plan.
The tools in detail
1. Shade
Shade is a cloud NAS built for video teams that want people search, review and cloud storage in one place. When assets are uploaded to a drive, images and videos in AI-indexed folders are scanned for faces, and unidentified faces show up in the asset's Details tab. Create a profile for a person from one of those faces, or add one in the People view with sample images, and Shade surfaces every place that person appears.
Names work inside AI search. A query like "JJ throwing a football" returns that person doing that action, not just any clip with that person. For better accuracy, Shade's docs recommend at least one clear, well-lit photo plus 2 to 3 extra samples. Shade can also trigger an automation when a person is recognized.
More detail is on the AI search page and in the release notes for the facial recognition update and Facial Recognition 2.0.
Pricing. Facial recognition is listed on both plans, not only the top one. Growth costs $29.75 per seat per month billed annually ($35 month to month) for up to 15 seats, each with 500 GB of active storage and unlimited AI indexing. Enterprise is quoted, with 1 TB per seat, SSO/SAML and SCIM. Add-on storage is priced on request (ask Shade for a quote), and there is a free trial.
Data and consent. You can exclude any folder from AI indexing, and faces in excluded folders are not scanned. Shade's AI policy and biometric FAQ say customers are responsible for notice and consent, workspace admins can delete biometric data by deleting individuals or the whole workspace, and customers are opted out of AI model training by default. The AI policy also says Shade uses third-party AI providers, including OpenAI and Gemini, for specific platform features, and that all AI hosting and processing of customer content takes place in the United States. Read both alongside the biometric data addendum.
Gaps to know. Indexing runs after upload, so new files may not appear right away. The People view lists only people you have named, not every detected face. Face data is per drive: faces indexed in one drive are not searchable from another.
2. Amazon Rekognition
Engineering teams building their own face search on AWS will start here. In Rekognition, you create a collection, index faces into it and search images or stored video against it. (AWS closed its streaming video analysis to new customers on April 30, 2026.) AWS says collections store face vectors, which are mathematical representations of faces, not images of faces. You can group up to 100 face vectors per person into a user vector to improve accuracy, though AWS describes user-vector search for faces from images, not video.
Pricing. Face search and face detection on stored video cost $0.10 per minute each (US East), and keeping faces in a collection costs $0.00001 per face per month. A new AWS account comes with up to $200 in Free Tier credits, which AWS says can be applied to Rekognition.
What you get back is matches, not a library. The interface, storage links and retention rules are yours to build, which suits products and systems more than editors.
3. Azure AI Video Indexer
Microsoft-based organizations that can meet Microsoft's approval criteria have a ready-made option. In Video Indexer, the Standard video preset adds facial recognition and celebrity recognition to the Basic preset (objects, labels, OCR, keyframes, shots) and costs $0.09 per input minute in US regions, with audio analysis billed separately.
The hurdle is access. Face identification, custom person models and celebrity recognition are Limited Access: you must register and be approved, access is limited to Microsoft-managed customers and partners for approved use cases, and trial accounts do not have access. Plan for lead time.
4. TwelveLabs
For developers who want to find specific people in video through an API, alongside search by meaning. Entity Search, which TwelveLabs labels beta, lets you create collections of people, upload several reference images per person and then find where each person appears in your videos. The same platform offers natural-language search and video analysis through its Marengo and Pegasus models.
Pricing. Free ($0), Plus ($29 per month) and Pro ($99 per month), each with set storage and usage allowances, plus custom Enterprise.
As an API, it leaves the library, permissions and review for you to build, and people search is still in beta.
5. iconik
A fit for teams already managing assets in iconik. Facial recognition, the iconik MCP server and bring-your-own AI services are on Professional and Enterprise only. Object and scene detection, speech to text and AI metadata are on all tiers.
Pricing. Face search needs Professional or Enterprise. Both are quoted and include monthly AI credits; iconik's FAQ puts that at 100 to 1,000 a month on Professional and 2,000 on Enterprise. Facial recognition on video uses 2 of those credits per hour of footage, but iconik does not publish what extra AI credits cost. The usage-based Starter plan ($0 to $120 per user per month by role) does not include facial recognition. See our iconik review.
Total cost depends on AI credits and on storage you buy separately.
6. Axle AI
Axle AI suits post teams that want face search on premises or in a private cloud. Its homepage describes "trainable faces," scene understanding, transcription, and logo and text recognition, and says AI tagging can run on premise, in a private cloud or in Axle AI Cloud. Axle AI Tags adds vector search and an API.
Pricing. Axle AI MAM from $2,995 (on-prem), Axle AI Cloud $20 per TB per month, Axle AI Tags from $200 per month. See our Axle AI review.
Expect more setup and administration than a pure cloud service. How well face search works depends on how you deploy and train it.
7. Moments Lab
For broadcasters, studios and sports teams that want character and talent search across a large catalog of shows. Moments Lab's MXT indexing describes who appears and what happens in each shot. You can train it to recognize your characters and on-screen talent by actor name and role, and it also transcribes speech and detects logos. It connects to on-prem and cloud storage, so originals can stay where they are.
Pricing. Quote-based; the site offers a demo rather than a price list.
It is aimed at catalog owners turning shows into social content more than at general production libraries. See our Moments Lab alternatives.
8. Immich (free, self-hosted)
Immich is open-source software (AGPL v3) for technical teams or individuals with a small, low-budget library and someone to run a server. It detects faces in photos and videos, converts each face to an embedding, clusters them with a density-based method, and lets you name, merge and search people. Face detection runs on each asset's generated preview image, so test how well it finds people in your video clips.
Pricing. Free to use; you pay for hosting and your own time.
It is built for photo and video collections, not professional production work. There is no review and approval, no permissions model designed for client work, and security, backups and updates are on you.
Privacy, consent and the law
Face data is treated as sensitive in many places. This is general information, not legal advice, so talk to counsel before you scan a library.
Illinois (BIPA). The Biometric Information Privacy Act regulates private entities that collect biometric identifiers, which include scans of face geometry. In general it calls for written notice, a written release from the person, and a published retention and destruction policy, and it lets individuals sue.
EU and UK (GDPR). Article 9 of the GDPR, and the same article of the UK GDPR, prohibits processing biometric data for the purpose of uniquely identifying a person unless an exception applies. One exception is explicit consent for one or more specified purposes. EU Member States can add further conditions for biometric data.
Other US states. Texas and Washington also have biometric privacy laws; Shade's biometric data addendum names both alongside BIPA.
Vendor rules. Microsoft restricts face identification through Limited Access. Shade's biometric FAQ says customers are responsible for notice and consent and for deciding how long biometric data is kept.
Before you name anyone, decide who you are allowed to identify (talent under contract, public figures, staff with consent). Keep a list of named people, set a retention rule and know how to delete face data. Then check what your vendor does with the data, whether it is used for training and where it is processed.
Picking the right tool for your library
If you need | Pick |
People search inside a cloud drive with review and storage | Shade |
People search inside an existing MAM | iconik (Professional or Enterprise) or Axle AI |
On-prem face search | Axle AI |
Faces sorted inside your editing app only | DaVinci Resolve Studio |
Character and talent search across a show catalog | Moments Lab |
A developer API with named matching | Amazon Rekognition, TwelveLabs (beta), or Azure if approved |
Free and self-hosted | Immich |
Test on your own footage first. Lighting, camera angles, masks, age changes and small faces in crowd shots all lower match quality, and no vendor on this list publishes numbers you can rely on for your material.
When Shade is the right call for people search
Shade suits teams that find a person and then need to do something with the footage: review it, share it and deliver it from the same place. You name a person once and combine the name with a description in full-sentence search. Security is SOC 2 Type II, ISO 27001 and TPN certified; HIPAA and GDPR compliant. See Shade pricing.
It is the wrong tool if you need face search on your own servers or a pure API for an in-house system. Teams of more than 15 seats need an Enterprise quote. And if your organization cannot use AI processing hosted in the United States, look at an on-prem tool or a cloud service that runs in your region.
FAQ
How much does face search cost on a cloud API?
By our estimate from list prices, searching 200 hours of stored video (12,000 minutes) against a face collection costs $1,200 on Amazon Rekognition at $0.10 per minute, and storing 5,000 faces adds $0.05 a month. Azure's Standard preset would cost $1,080 for the same footage, if Microsoft approves your use. Per-seat platforms such as Shade include AI indexing in the plan instead.
How many photos does face search need per person?
More good samples help. Shade suggests at least one clear, well-lit photo and two or three more, and TwelveLabs' Entity Search takes several reference images per person. AWS can pool up to 100 face vectors per person into a user vector, though it documents that for searching images, not video.
Do I need consent to run face recognition on my archive?
Often, yes. Illinois BIPA calls for a written release, and the GDPR allows biometric identification on explicit consent or another listed exception. Your vendor's terms matter too: Shade's biometric FAQ makes notice and consent the customer's responsibility. This is not legal advice; ask counsel.
Is there free face recognition software for video?
Immich is free, open-source and self-hosted, and it recognizes faces in photos and videos. On the cloud side, the Free Tier credits AWS gives new accounts can go toward Rekognition face search. Both need technical setup and offer fewer team features than a paid platform.
Does Shade store face data?
Yes, for customers who use facial recognition. Shade scans faces in AI-indexed folders to support search, keeps face data scoped to each drive, and lets workspace admins delete biometric data by removing individuals or deleting the workspace. Its AI policy says face clusters are encrypted at rest and in transit and stored with strict tenant isolation. Customers are responsible for notice and consent. See Shade's biometric FAQ for details.