I’m gonna set you straight: Tracking is for suckers. Stabilizing is for winners.
Let’s say you’ve got a label you need to stick on the side of a product. You could track the label onto the product. It’s not wrong; a lot of sucker-type activities are not wrong. What all sucker-type activities have in common, however, is they ignore easier and better methods.
Be a hero: stabilize it.
Sticking a graphic onto a still object is easier than sticking a logo onto a moving object. Adding lighting is easier, tracker removal is easier, paint is easier. Just about everything is easy as heck if you’ve got a stabilized image. It’s super nice to be able to see if (and how) your added graphic is wiggling or not sticking, which can be very hard if that image is moving around the screen.
WAYS TO STABILIZE
- One-point track
- Two-point track
- Planar track
- Auto-Stabilize
METHOD 1: the one-point track.
Load an image into Action, go to the axis’s “Tracking” tab and enter the tracker. Track whatever bit you want to stabilize, then exit out and activate the “Invert” button to the right of the axis’s position/rotation/scale/shear/pivot controls.
This works similarly to the old “stabilize” option, but unlike “stabilize” which is hacky as all get out and splits your transform data across and axis and image node, this inverted-track has all of the transform data in one place. It is also dead easy to un-stabilize, which I’ll get to. To summarize: don’t use the “stabilize” option to stabilize. Track w/ the Invert button on is better.
Now you have the element of your track behaving as though someone stuck a pin in it. Congratulations: your element is stable.
Now, should you want to reposition this stabilized element into the center of the screen to maximize real-estate, please do so by adding an axis above the inverted track. This is the opposite of how you work when tracking (vs stabilizing), but becomes crucial when it’s time to un-stabilize your work.
(the incorrect and correct ways to order your offset axis when stabilizing)
METHOD 2: the two-point track.
Just like method one, but you activate “Rotation” and/or “Scale” (depending on what you need to stabilize the shot) and run both trackers. The red tracker is your primary one, so place it on the area you want stabilized. The green tracker will figure out rotation and scale and should be placed as far away as you can while still being attached to the object to be stabilized. Your scale and rotation noise increase with tracker proximity.
Both trackers run at the same time by default. This can be a pain when a tracker needs to be manually managed to run properly. While you can temporarily turn off the difficult tracker, an easier way is to change “Selected” toggle to “Solo“ (shout out to the wonderful @kirk for this, I think of you every time I flip the toggle, still). Solo will only run the tracker that you have active and will behave as though all other tracks are off.
[
]Once you’ve got your two trackers tracked, the steps are the same: exit the module, turn on “Invert” on the tracked axis and add an offset axis if necessary.
METHOD 3: planar tracking.
Planar tracking is nice because you don’t have to jump into a separate tracking module. You can do it all right in Action (or GmaskTracer, which has the same tools). Draw an action gmask around the thing you want to track. Keep anything that you think is going to upset the track outside and anything that will help inside. For my track of this asthma inhaler I made a mask that looks like this:
No need to get fancy on it. Leave all the gmask settings as they are–don’t invert it or anything. Double click on the axis above the gmask and go to the Tracking tab in the Object menu, but this time select “Planar”. This will bring up a whole bunch of options that should be familiar enough, tracking-wise. (Track forward, track back, auto-update ref, set new ref, etc…) and you hit go and hope for the best. If you want to see what the track is seeing and comparing it to, you can run the track in F8 View, which will show you thumbnails of your reference and what the tracker is seeing on the current frame.
You don’t have to use the F8 View, but it’s helpful if your track goes wonky.
The planar tracker has two modes: “region warping” and “multi-feature detection”. If region warping doesn’t work, try multi-feature detection; it may work better.
And now you have your planar track! Copy the axis with the track data, hide the gmask, and parent your image to the copied gmask track. Then hit the invert button (of course). Offset if necessary.
METHOD 4: Auto-Stabilize.
It does what it says on the box, and it’s often pretty good. I use it for fullscreen stabilization–removing camera twitches or gate weave. The three preceding methods are better for object-focused stabilization.
The first important thing with Auto-Stab is that you turn on “Hardware Filtering”.
Hardware filtering is great. You should always use it. Failure to do so will result in softening your image a lot. Hardware filtering is so good it’s almost magic. Half the reason I wrote this tutorial is to tell you about it.
The second important thing to know is that you should set your start frame on the node to one frame lower than you want. I don’t know why this is, but if you leave it at frame 1 and track, the first frame will not be tracked, so set it to zero, or whatever is one less of your start frame.
TIME TO UNSTABILIZE!
Not much point to all this work if you can’t get your static element back into the plate easily.
It’s easy:
Copy the action that is stabilizing the image, feed in your newly painted up, stable image and replace the image under the inverted axis (which may happen automatically if they’re off the same layer input) and turn off “invert”. If you have an offset above it, press the INVERT button, move the offset BELOW your stabilizing node (anything done must be undone in the reverse order. Your stabilized comp will now track with your source image! You’ll probably need to gmask the comp back over the plate, but because your image is stabilized it’s easy to do that–just gmask the stabilized image before going into the un-stabilizing action. Often you won’t need to even set keyframes. Yet another win for the hero-type behavior of stabilization.
You can stack up stabilizations, using one to take most of the move out and a second to remove the rest of it. I usually build this by making sucessive actions, and then copying each new sub-stabilization into ANOTHER action so they can all concatenate and only cause one filter hit. Make sure to de-stabilize in reverse order–taking off the second stabilize before taking off the first (this too can all be done in a single action).
To invert an Auto Stabilize activate the “negate stabilization” button. Make sure Hardware Filtering is still on.
And that’s it.
in short, stabilize by using the “track” function and inverting the axis. Unstabilize by turning off the invert. There’s a lot of variation to how this can be approached, but it’s dead handy and the core of my personal flame toolset.
For further stabilization, check out this cool method @kirk made:
https://forum.logik.tv/t/not-randys-black-belt-camera-projection/