Dance and performance
Let audiences watch choreography from the angle they choose; keep rehearsals and repertoire as a moving 3D archive.
4D Gaussian Splatting platform
Turn your multi-camera shoot into
a moment you can watch from any angle.
Upload your videos and 4dgspl.at rebuilds your moving scene in 3D and over time. Viewers turn the camera wherever they like while it plays.
Behind the scenes
How does it work?
See what happens step by step, from the moment your videos are uploaded to the share link. Hover over or tap a step. In short: multi-camera video to 4DGS or Blender render to 4DGS.
Multi-camera capture
Cameras placed around the scene record the same moment from different angles. The more angles, the more complete the 3D scene.
Videos started in sync · same frame rate
Upload to the site
Create a project in your dashboard and drop each camera's video into its slot. Files go straight from your browser over an encrypted connection.
4dgspl.at · project page
Cloud: storage and job queue
Your videos sit in private storage that only you and our team can access. Once you submit, the project enters the processing queue.
Supabase · private storage · row-level access rules
Processing server
The worker on our GPU server picks up the next job from the queue by itself, then downloads and checks the videos. No inbound connection to the server is ever opened.
NVIDIA RTX 5080 · restricted worker account
Camera calibration
Each camera's position, orientation and field of view are determined. Without a calibration file, COLMAP finds feature points in each camera's frame from the first second, matches the same points across cameras, builds a sparse point cloud and solves the camera poses.
COLMAP: features → matching → sparse model → camera poses · or your own cameras.txt + images.txt · automatic COLMAP is being installed on the server
4D Gaussian engine
Each frame is rebuilt from thousands of small, semi-transparent, colored "Gaussians"; the engine trains them on the GPU until they match what every camera saw.
gsplat (Apache-2.0) · frame-by-frame 3D Gaussian Splatting
3D player in the browser
The result is ready on your project page: rotate, zoom and pause the scene while it plays. No app to install.
PlayCanvas · per-frame compressed.ply
Share link
Create a link that needs no sign-in with one click and share the result with anyone. You can close the link at any time.
Public link · input videos are never shared
data flow the player reads results from the cloud through time-limited signed URLs
1 / 8
Multi-camera capture
You control the camera
4DGS content plays like a video, but viewers can change the viewpoint freely while it plays. Drag to rotate, scroll or pinch to zoom.
about 8.7 MB
0:00.0 / 0:00.0
Synthetic scene — not a real capture Pollen Robotics' open-source Microduck robot was animated dancing, jumping, falling and getting up, rendered in Blender from 8 virtual cameras at two heights and processed by the 4dgspl.at engine (60 frames, 15 fps). Make your own →
Microduck © Pollen Robotics; the 3D model and this sample scene are CC BY-NC-SA 4.0 (rendered and converted to 4DGS). Pollen Robotics does not endorse this work.
Use cases
Where does it help?
Wherever the motion itself matters and viewers wish they could see it from over there, too.
Motion analysis
Freeze a jump, a strike or a technique and orbit around it; coach and athlete study the same moment from every side.
Living heritage
Document a craftsperson's hands, a folk dance or a lab experiment over time and from every angle.
Ads, music video, XR
Show products and artists on your site with a free camera; bring scenes made with a virtual rig into game engines.
Workflow
From capture to publishing, step by step
Every project follows the same six steps. At each one you can see what you do, what we take care of and what you get.
-
01
Plan the shoot
Start with a short checklist for camera count and placement, lighting and background.
- You
- Choose the scene and the performer; place the cameras around the performer.
- Us
- We share a pre-shoot checklist and a camera placement suggestion, and review the setup with you if needed.
- Output
- A setup ready to shoot, with recording settings (fixed frame rate, fixed exposure).
-
02
Shoot
All cameras record the same moment. Pick the setup that suits your project.
- You
- Give a sync cue at the start of the take (clap, flash or timecode) and record a short calibration clip.
- Us
- We tell you which cameras and phones work well and which settings to use.
- Output
- One video file per camera.
Option 1 Fixed studio rig
A permanently installed, calibrated multi-camera rig with fixed lighting. Consistent results across sessions.
Option 2 Portable setup
Cameras or smartphones on tripods. Take it to set and have it running in minutes.
Option 3 Virtual rig
4–8 virtual cameras in Blender, Unreal or Unity; exact calibration, and AI coding agents can build it for you.
-
03
Upload
Create a project in the dashboard and drop each camera's video into its slot.
- You
- Upload the videos and the calibration (or create it with COLMAP), then press "Send for processing".
- Us
- As soon as the files arrive we check that duration, resolution and frame rate match.
- Output
- A project marked "Queued" in your dashboard.
-
04
We process it
The cameras are aligned, the performer is separated from the background and the scene is rebuilt over time with 3D Gaussians.
- You
- Nothing; follow the status in your dashboard.
- Us
- We run the stages below on our GPU infrastructure; if something goes wrong, we explain why in the dashboard.
- Output
- A 4DGS sequence in which every frame is a 3D Gaussian scene.
- File check
- Time sync
- Calibration (COLMAP)
- Performer mask
- 4DGS training
- Compression
Draft→ Queued→ Processing→ Ready -
05
Review
Play the result in the browser, pause it and look from any angle.
- You
- Scrub the timeline in the player and rotate the camera freely.
- Us
- We run quality control and reprocess if needed.
- Output
- Approved content, ready to publish.
-
06
Publish or export
Deliver the content to your audience or use it in your own tools.
- You
- Share the link, embed it on your site or download the files.
- Us
- We prepare a compressed, streaming-friendly package for the web.
- Output
- An embeddable player and downloadable files.
- Web player — share by link or embed on your site
- Per-frame 3DGS — download as a PLY / compressed PLY sequence
- Game engines and SDK Coming soon
What you get
Free-viewpoint playback
Viewers rotate the camera, zoom and pause at any moment while it plays.
Runs in the browser
No app to install; it opens in desktop and mobile browsers and embeds on your site.
Open file format
Standard per-frame 3D Gaussian files that you can also use in your own tools.
Who it's for
Studios
Process the data from your volumetric capture rig without building your own infrastructure.
Production companies
Add a new way to watch commercials, music videos and stage performances.
Volumetric content creators
Produce dynamic scene assets for XR, the web and game engines.
Frequently asked questions
What is 4D Gaussian Splatting (4DGS)?
4D Gaussian Splatting represents a moving scene with a large number of small, semi-transparent, colored 3D "Gaussians" that change over time. It is learned from videos recorded simultaneously by several cameras. The result is a 3D video that plays in real time in the browser and that viewers can watch from any angle. At 4dgspl.at each frame is produced as its own 3D Gaussian scene, and the player shows them in sequence.
How is it different from 360° video and classic volumetric video?
360° video lets you look around from a single point; the viewer cannot move. With 4DGS the scene itself is rebuilt in 3D: the viewer can circle around it, move closer and look from different angles. Because it uses Gaussians instead of the meshes or point clouds of classic volumetric video, hair, fabric and semi-transparent details usually look more natural, and the scene plays smoothly in the browser.
How many cameras do I need?
We recommend 8 cameras, with 6 as the minimum. Four cameras also process, but views from angles between the cameras show ghosting. In our tests 16 cameras made no clear difference over 8 at today's processing resolution. With 8 or more cameras use two heights, and let the performer fill 70–85 % of the frame height.
Can I shoot with smartphones?
Yes. Phones need a fixed frame rate, fixed exposure and fixed focus; the capture guide explains step by step which settings to choose and how to sync the recordings.
What is camera calibration (COLMAP) and how do I create it?
Calibration describes each camera's position, orientation and field of view in the scene; without it the engine cannot combine the views. You create cameras.txt and images.txt with the open-source COLMAP from one frame per camera taken at the calibration moment and upload them as a zip; our Blender script writes them directly from virtual cameras. Automatic estimation from the images on the server is in development for uploads without calibration. The steps are in the capture guide.
Can I create 4DGS content with AI tools?
Yes. Coding agents such as Claude Code or Codex can run Blender headless, build the model you describe, render synchronized videos from 4–8 virtual cameras and write the COLMAP calibration; you then upload the result to 4dgspl.at. The window in the capture guide generates a detailed prompt for this. Text-to-video models are not suitable because they cannot keep the views geometrically consistent.
Do you shoot, or only process?
We currently offer processing: you or your studio does the shoot and uploads the files. We help you set up the capture rig.
How long does processing take?
It depends on the number of cameras, the length of the take and the resolution. You can follow the status step by step in your project dashboard.
What formats do you deliver?
A shareable, embeddable player that runs in the browser, plus per-frame 3D Gaussian files (PLY / compressed PLY). Plugins for game engines are in preparation.
Can we try it first?
Yes. Sign up with email or Google; every new account starts with 5 free jobs of up to 10 seconds each at Fast quality. After that you use credits; credit packs are coming soon.
Get started
Create an account with email or Google and upload your first multi-camera shoot. Every new account starts with 5 free jobs (up to 10 seconds each, Fast quality).