AI camera tracking. The solve CG waits on, back in a day.
No CG element can go into a shot until someone works out exactly how the real camera moved. A matchmove department does that by hand, placing and nursing tracks frame by frame, and every 3D shot on the show queues behind it. Our AI tracks the plate and solves the camera, focal length and lens distortion in one pass, and a matchmove artist checks the solve before it leaves, so your CG team starts tomorrow instead of next month.
Houdini · Blender
What a matchmove department tracks by hand, the model solves in one pass
We are not a VFX lab with faster artists. We are an AI lab: the tedious, per-frame work that fills a vendor’s schedule is done by the model in a single pass, and artists are spent only where judgment matters. The result is the same deliverable, in a day or two instead of weeks.
| The job | Traditional VFX lab | Pixazo AI |
|---|---|---|
| Tracking features | Tracks placed, nursed and cleaned up by hand | Features found and tracked across the whole plate by the model |
| Solving the camera | Iterated by an artist until the error is low enough | Camera, focal length and lens distortion solved together |
| Turnaround | Weeks per sequence, and every CG shot waits on it | 24–48h on a standard shot |
| Cost driver | Artist-hours per shot | Shot class; one free test shot gives the number |
Where the artist still matters: solves that drift, shots with too little parallax, and setting scale and orientation from the set survey. Every solve is checked before it leaves, and the solve error is reported per shot.
From plate to a camera your CG team can use
The deliverable is not a track, it is a camera: something you open in your 3D package and look through, with the CG lined up to the plate on every frame.
What actually comes back
Tell us which package the shot goes to next and the files arrive ready to open there, in your naming convention.
| Deliverable | Format | Notes |
|---|---|---|
| Solved camera | .nk, .ma, .hip, .blend, or FBX / Alembic | Animated camera with focal length, film back and frame range matched to the plate. Opens in Nuke, Maya, Houdini or Blender. |
| Point cloud | Locators in the scene file, or Alembic / OBJ | The tracked features in 3D, plus simple reference geometry for the ground plane and key surfaces so layout can start straight away. |
| Lens distortion | Nuke LensDistortion node, or ST maps as EXR | Undistort and redistort, so CG is rendered on straight lines and put back through the real lens in comp. |
| Scene scale & orientation | In the camera file | Set from your survey or measurements; otherwise from known objects in the plate, with the reference used noted in the delivery. |
| QC render | ProRes or H.264 | The plate with a checkerboard and test geometry locked to it, so you can see the solve hold before anyone builds on it. |
| Object tracks (on request) | Transforms in the same scene | Rigid objects that move in the shot, such as a car, a prop or a screen, tracked for CG attachment. |
The question that decides the scope: does the camera translate? A camera that moves through space gives the solve parallax to work from and is a standard shot. A locked-off or purely panning camera is solved as a nodal pan, which is quick but gives no depth, and body or face tracks for performance are a different job.
AI camera tracking cannot solve every shot
A solve needs detail that stays still while the camera moves. Where the plate gives it that, the pass is fast and accurate; where it does not, no tracker can invent it.
- Moving cameras on textured sets. Dolly, crane, Steadicam, drone and handheld over locations with real detail: streets, interiors, landscapes.
- Plates with people in them. Moving people, traffic and water are ignored by the solve rather than tracked by mistake.
- Known and unknown lenses. With a camera report the lens is used as given; without one it is solved from the plate.
- Long shots. Oners and long drone moves are solved as one camera, not stitched from pieces.
- Sequences on the same setup. Shots from one lens and location share a lens solve, so they line up with each other as well as the plate.
- Green or blue screen with no markers. A flat, featureless wall has nothing to track. Markers on set, or a survey of the stage, fix it.
- Heavy motion blur and rolling shutter. Whip pans and fast handheld on rolling-shutter cameras are solved with an artist, frame range by frame range.
- Zooms with no lens data. A zoom during the shot without the focal length recorded leaves focal length and camera move ambiguous.
- Nodal pans needing depth. A camera that only rotates gives no parallax, so the solve is a pan with no 3D layout behind it.
- Body and face tracking. Performance tracks for digital doubles and face replacement are scoped separately.
What helps most on set: the camera report (body, lens, focal length, filmback), a few measurements between things that do not move, and markers on any flat screen. With those, almost every shot is a one-pass solve.
What moves the number
We do not publish a rate card, because a slow dolly through a textured street and a whip-panning handheld shot on a green screen are not the same job. These are the drivers; the test shot gives you the actual figure.
| Driver | Cheaper | Dearer |
|---|---|---|
| Camera move | Smooth dolly, crane, drone, Steadicam | Fast handheld, whip pans, rolling shutter |
| Set | Textured locations and dressed sets | Featureless screens without markers |
| Lens | Prime with a camera report | Zoom during the shot, no lens data |
| Deliverables | Camera and lens data | Object tracks, survey-matched scale, multiple packages |
| Volume | A sequence or a show, priced per shot class | One-off shots |
Faster than a rate card: send one shot, ideally the one your CG team is waiting on. We return a solved camera with a QC render and a firm per-shot number for that class of work. If you are already being quoted elsewhere, tell us the number you need to land under.
Frequently asked questions
What is AI camera tracking?
Working out exactly how the real camera moved, along with its focal length and lens distortion, so CG can sit in the shot. The AI tracks and solves in one pass; a matchmove artist checks the solve.
Which software can I open the solve in?
Nuke, Maya, Houdini and Blender scene files, or FBX and Alembic for anything else, with lens distortion as a Nuke node or ST maps.
Do you need lens and camera data?
It helps. With a camera report the lens is used as given; without one it is solved from the plate. A zoom during the shot with no focal length recorded is the hard case.
Can you track green-screen shots?
Only with markers on the screen or a survey of the stage. A flat, featureless wall has nothing to track.
How long does it take and what does it cost?
24–48 hours on a standard shot, with the solve error reported per shot. We do not publish a rate card. Send one shot free and you get a finished result plus a firm per-shot number for that class of work.
Roto, paint and matchmove, all AI
Matchmove is the third leg of the work studios send out, alongside AI rotoscoping and AI cleanup and paint. Once the camera is solved, a clean plate or an AI crowd can sit in the same shot. All of it is inside the Pixazo AI Lab: 15 production capabilities, 200+ models, human-finished on every shot.
Send us the shot your CG team is waiting on.
One shot, no charge, no commitment. You get a solved camera, the lens data and a QC render, plus a firm number, and your layout artist opens the scene before you decide anything.
Send a test shotCamera solve · Point cloud · Lens distortion · 24–48h standard shot · Artist-checked
