Perfect annotations by construction
Every label comes from the scene itself, not from a human guess. Pixel-exact masks, metric depth and exact poses on every frame.
For AI labs
Custom 3D training data, built to your spec.
Photorealistic 3D scenes and animation with perfect ground truth - depth, segmentation, camera and physics parameters included with every frame.
Every engagement starts with a pilot batch.
Why 3D data from a studio
Real-world capture gives you footage and a labeling problem. A built scene gives you both the image and the exact truth behind it.
Controlled variation
Change lighting, camera, materials and environments one variable at a time, so you know what your model is learning from.
Edge cases on demand
Rare events, unusual objects and long-tail conditions, built deliberately instead of waited for.
No licensing or privacy constraints
Every asset is created for your dataset. No bystanders, no scraped footage, no rights to clear.
Impossible to capture
Scenarios that are dangerous, rare or physically impossible to film: collisions, failures, extreme conditions, extreme scales.
Production-grade realism
Built by a professional 3D animation team: physically based materials, real camera optics, believable motion.
What you get
Every frame ships with its ground truth, rendered from the same scene, so labels and pixels always agree.
- Rendered sequences
- RGB frames or video at your resolution, frame rate and aspect ratio.
- Depth maps
- Metric depth, per pixel, for every frame.
- Segmentation masks
- Semantic and instance masks, per frame, with your class list.
- Surface normals
- Per pixel, in camera or world space.
- Camera parameters
- Intrinsics and extrinsics for every frame, including lens and motion.
- Object trajectories
- 6-DoF poses over time for every tracked object.
- Physics parameters
- Masses, velocities, forces and contacts where the scene is simulated.
- Structured metadata
- Scene, lighting, material and render settings for every sample, machine-readable.
Delivered in your format, with a QA pass on every batch before it reaches you.
Use cases
Built for teams whose models need to understand space, motion and physics.
World models
Long, physically consistent sequences with known camera motion and object dynamics.
Robotics and manipulation
Grasping, contact and clutter scenes with exact poses and physics, so policies learn from precise labels.
Generative 3D
Clean geometry, materials and multi-view renders with matching ground truth.
Video generation
Controlled motion, camera moves and lighting, with every parameter recorded.
Computer vision
Detection, segmentation, depth and tracking data for the cases your real-world data misses.
How it works
From spec to production, with your approval before anything scales.
- Send us your dataset spec.
What the model needs to learn, the scenes and variation you need, annotations, volume and format.
- We build a pilot batch.
A small, representative batch built to your spec, delivered for your review.
- You approve quality and annotations.
Test it in your pipeline. We adjust realism, variation or labels until it meets your bar.
- We scale production.
Full volume, delivered in batches on your schedule, each with a QA pass.
Commission a dataset
Tell us what your model needs to learn. We reply within one business day to scope a pilot.
- NDA first. We sign yours before you share a spec.
- Yours alone. Full rights on delivery, never reused across buyers.
- Pilot before scale. Approve quality and annotations first.
Brief received - we reply within one business day.
Questions in the meantime: [email protected].
Questions
Who owns the data?
You do. Full rights to the delivered dataset transfer to you on delivery, so you can train, evaluate and ship models on it without restrictions from us.
Can you work under NDA?
Yes. We sign your NDA before you share a spec, and we keep the project, your requirements and the dataset confidential.
Do you reuse client assets or scenes across buyers?
Never. Every dataset is built for one buyer. Scenes, assets and renders made for you are not reused, resold or shown to anyone else.
Can we start with a pilot?
Yes. Every engagement starts with a paid pilot batch, so you can check realism, annotations and format against your pipeline before we scale.
What formats do you deliver?
Yours. We match your pipeline: image and video formats, annotation schemas, folder structure and metadata, agreed during the pilot.