Custom human video datasets for AI

Turn a model data gap into an accepted human-video pilot.

We design the protocol, capture real participants with documented permissions, QA the material against agreed acceptance rules, and deliver structured media — not loose footage.

Describe what your model needs to see See real capture examples

Participant looking to the side during a controlled facial-motion capture — Face & speech

Hand tracing a colored shape with a 3D pen — Hands & objects

Adult participant adjusting her jacket in a controlled studio task — Human actions

A produce-cutting task shown from chest-mounted and third-person cameras — Paired views

Real participantsDefined protocolsDocumented permissionsAcceptance-led QAPrivate delivery

Proven capture archetypes

Start from a real production pattern — then adapt it to your task.

These public examples show capture patterns already produced by Continental People. Project feasibility, participant scope and intended use are confirmed during review.

Participant forming a rounded lip shape during a controlled speech prompt

Face & speech

Face and audiovisual speech

Repeated prompts, controlled conditions and synchronized speech for face-centric research and evaluation.

See FaceMotion-16

Procedural capture

Hands, objects and procedures

Fine hand movements, object interaction and repeatable task steps for procedural and physical-world workflows.

See Creative Hands See Everyday Dexterity

A packaging task shown from chest-mounted and third-person cameras

Multi-view events

Human actions from multiple viewpoints

The same event can be captured from egocentric and third-person views when the protocol requires it.

See the paired-view demo

Adult participant moving her hair during a controlled grooming task

Supporting pattern

Controlled body & presenter performance

A published proof-of-concept for controlled appearance, body motion and performed prompts.

Review Studio Humans

Project-specific capture

Custom capture with young participants

Projects with young participants are discussed individually. Age ranges, tasks, capture formats and participation conditions are agreed for each project.

The video shows one example: a capture setup for age-assurance and presentation-attack research, featuring live capture, printed photographs and screen-based presentations. Other projects are scoped around your specific requirements.

Discuss your project

Descriptive transcript

A single 15-year-old participant is recorded in four presentation states: live, a printed photograph, a photograph displayed on a screen, and a video replay displayed on a screen. Setup shots show the camera and presentation materials. The film includes music and short participant speech. It demonstrates a capture setup, not an age-estimation or attack-detection result. The final invitation is to discuss a project-specific capture brief.

From data gap to acceptance

Review the pilot before committing to scale.

Define the data gap

Clarify what the model needs to see, the people and actions involved, the required views and the material constraints.

Approve the pilot specification

Agree the capture matrix, participant profile, metadata, permissions and acceptance criteria before production.

Review capture and QA

Evaluate the pilot material and documented QA results before a larger collection is planned.

Expand or stop with a usable package

Freeze the accepted version for delivery, or define a funded expansion from the reviewed evidence.

What accepted delivery looks like

A reviewable dataset package, not a folder of loose footage.

The exact package is agreed per project. Approved media stays connected to its capture plan, metadata, QA records, version and delivery documentation.

Captured media and agreed views
Structured metadata fields
Accepted, rejected or limitation status
Versioned package anatomy
Manifest and checksums
Dataset Card, QA Summary and known limitations

Illustrative structureVersioned package

captured_media / agreed_views

capture_event
camera_view
action_or_prompt
capture_condition

metadata + QA status

acceptance_status
limitation_note
package_version

delivery documentation

manifest + checksums
Dataset Card
QA Summary

Real delivery example · Household E021

One orange-cutting event, two linked views

These public 1080p excerpts show chest-mounted and fixed third-person views of the same physical event. Their playback times are not a claim of frame-level synchronization.

Chest-mounted view · public web preview

Third-person view · public web preview

Related metadata excerpt

Event: Household E021
Task: Cutting an orange with a kitchen knife
Views: Chest-mounted and fixed third-person
Capture mode: Paired physical event

The underlying 4K evaluation sidecars describe separate silent H.264 representations, not the public previews shown here.

4K chest-view sidecar excerpt: 3840 × 2160 · 12.604 s · no audio · SHA-256 2d3e53e954c9f6502d6cf507350e01cad220ab7140897b83fcc2afc32571fc85. The 4K media is not publicly downloadable.

Web-preview manifest excerpt

SHA-256 checksums for the two public 1080p files, not the private 4K derivatives:

Chest · 13.12 s · 9,318,991 bytes: c927221e05a2887115fcd58c98795da505e15630354c51409007514c792f55af
Third-person · 13.013 s · 2,745,755 bytes: 55691c50d36267c41d4ddf6ce04db89d2d529c50ecdc023540d687a4a63170a0

Mika, Continental People's AI-generated virtual presenter

41-second overview

See the complete capture approach in 41 seconds

Mika is our AI-generated virtual presenter. Dataset footage features real people.

Watch on the homepage

Start with the essentials

What should your model see or understand?

A rough data gap is enough for a first review. If the project fits our production model, we will follow up to define the pilot, acceptance criteria and commercial scope.

This is a feasibility request, not an order or production commitment.

.js-request-form{display:none!important}This request form needs JavaScript. If it is unavailable, please contact Continental People by email and we will help with your request.

Start with the buyer need. Camera, annotation and sensor details can follow after the first fit review.

Full name *Work email *Company / organization *Country *

What should your model see or understand? *

Approximate scaleDesired timeline

Company website or specification link

I have read the Continental People Privacy Notice v2.4 and understand how my information will be used to process and protect this request.I confirm that I am authorized to submit this request and act for the organization named above.I agree that Continental People may contact me about relevant visual data collection services. I can unsubscribe at any time.

Protected by Cloudflare Turnstile for security and abuse prevention. See our Privacy Notice.

Boundaries & next steps

Questions before a brief

Can we start with a pilot?Toggle

Yes. A bounded pilot can be scoped so your team can review capture, metadata and agreed acceptance evidence before any expansion. Feasibility, price and timing are confirmed after review.

Which participant profiles and locations can you support?Toggle

Participant scope and geography are reviewed against the project specification, permissions and production feasibility. Additional locations can be assessed for a defined funded project.

Can annotations or additional sensors be included?Toggle

Potential annotation, camera, audio or sensor requirements are assessed during technical scoping. Submission does not confirm availability or inclusion.

How are participant permissions and intended uses handled?Toggle

Permissions and intended uses are scoped per project and reflected in the applicable agreements. We do not promise unrestricted use, participant identities or release scans.

What happens after we submit a brief?Toggle

We review production fit and may request more detail for a pilot scope. A submission is not an order, quote or commitment.

Custom capture

Turn the requirement into a pilot your team can review.

Start the brief Use the detailed buyer brief