Workflows

Object Tracking

Solve a moving object and the camera separately, then bring them together in one scene.

When something in your shot moves independently, one solve can't describe it. The camera moves, the object moves, and the tracker has no way to tell those apart.

So you solve them separately. Mask the object out and solve the camera from the background. Then use only the object and solve that. Finally combine the two, set the scale, and align your mesh.

What This Looks Like

Import as Footage -> AI Mask ─┬─> Auto Track (Exclude)  -> the camera
                              └─> Auto Track (Limit to) -> the object
                                                    │
                       Manual Tracks ───┐           │
                                        ▼           ▼
                                     Object Solve
                                        │
                                     Output 3D

One mask feeds both solves. There's no inverted second mask to make.

Step 1: Mask The Object

Add AI Mask and prompt for whatever is moving. car, person, whatever it is. See AI Mask if you haven't set it up yet.

Dilate the mask by 5 to 15 pixels. The edge pixels belong to the object, and they're exactly the sort of high-contrast detail a tracker latches onto, so a mask that stops right at the silhouette still leaks.

Generate the sequence.

Step 2: Solve The Camera

Add Auto Track. Connect the image sequence, and connect the AI Mask into Mask Frames.

Set Mask Mode to Exclude Masked Area.

Now the solver ignores the object entirely and sees only the static world. Press Auto Track. This is your scene solve, and everything else is measured against it.

Check it before moving on. If this solve is bad, nothing downstream can be good.

Step 3: Solve The Object

Add a second Auto Track. Same image sequence, same mask.

This time set Mask Mode to Limit to Masked Area.

Now it's the other way around: only pixels inside the mask are used, so the solver works purely on the object. Press Auto Track.

Nothing needs inverting. The mode flips how the mask is read.

Step 4: Combine Them

Add Object Solve. It takes both:

• Scene gets the camera solve from step 2
• Object gets the object solve from step 3

Give it an Object Name, then press Solve Object. It works out the object's motion relative to the solved camera and outputs Object Data.

Useful settings:

When the solve finishes the node reports how many frames and points it used, plus the object error in pixels.

Step 5: Set The Scale

Here's the thing about solving an object separately: the result is ambiguous. A small object close to camera and a large one further away produce exactly the same picture. Nothing in the footage can tell you which it is.

Object Scale on the Object Solve node resolves it. Dragging it scales the object and moves it toward or away from the camera at the same time, so the screen match never changes. What changes is where it sits in your scene.

Drag it until the object sits at a believable depth against your point cloud. It updates live once the object is in the scene, so wire up Output 3D, press Import to Scene, then adjust while watching the viewport.

Step 6: Align Your Mesh

The solve gives you an empty with the right motion. Getting your actual model onto it takes a few hand-placed points.

First, hand-track some recognizable features on the object in the Movie Clip Editor. Corners, badges, bolt heads, anything you can identify on your model too.

Feed those into Object Solve's Tracks input as Manual Tracks and press Re-Solve. This matters: the linking step can only use tracks the object solve actually carried through. Link tracks that weren't wired in and the node tells you they aren't part of the solve.

Then link them to your mesh:

1. In the Movie Clip Editor sidebar, open 2D/3D Constraints.
2. In Constrain To, press + and add a target named for your solved object.
3. Select a track, put your mesh in Edit Mode, select the matching vertex, press Link to Track.
4. Repeat for the rest.
5. On the 2D/3D Constraints node, pick the object and press Align to Object.

Full detail in 2D/3D Constraints.

There's also a Lineup Object picker on Object Solve. Set it to your mesh and it gets parented under the solved empty automatically at import, which is often enough on its own if the object's origin already matches.

When It Doesn't Work

The object solve has too few points. Small or low-detail objects don't give the solver much. Hand-track features and wire them into Tracks, which works alongside the automatic solve rather than replacing it.

The object drifts against the plate. Check the scene solve first. Object motion is measured relative to the camera, so camera error shows up as object error.

The mask is leaking background into the object solve. Raise the dilate value, or tighten the prompt so the mask hugs the object more closely.

"Target is not the solved object." The name in the Constrain To dropdown has to match the Object Name on the Object Solve node.

"None of the linked tracks are in this object's solve." The hand tracks weren't wired into Tracks when you solved. Connect them and press Re-Solve.