Drone maker Skydio has a new automated shot design feature that should have filmmakers paying attention.
In the world of filmmaking, the drone market has been dominated by a single name, DJI. While contenders like GoPro and Parrot have tried to work their way in, DJI has remained at the top of the heap.
Skydio, best known among action and sports shooting enthusiasts for their excellent object tracking and self-flying features, are now out with a feature that is aimed squarely at filmmakers, Keyframe.
Keyframe is being released alongside the brand new Skydio 2+ drone, but in a nice moment of corporate goodwill, Keyframe will also be available on the normal Skydio 2. Because of schedules in testing, we only reviewed the feature on the Skydio 2.
Before we dig into Keyframe in particular, let's talk about Skydio for a moment. Their whole thing is amazing autonomy. There are some great comparison videos online showing how much better Skydio is at following subjects than the competition, especially through thickets of trees and other obstructions.
It does this by covering the drone with cameras, then using its Autonomy Engine to build a 3D model of the world that the drone can use to navigate.
The tech is truly astounding. If you've flown any modern drone, "return to home" is a common feature, but it's way more powerful with a drone like the Skydio 2 since that return to home can fly around things in the way. If you want "return to home" to work on most drones, you need a completely clear field. With the Skydio, it can fly through trees in a way that is absolutely astounding to watch.
In fact, on one occasion it tried to fly through some trees and discovered it couldn't. It asked me to fly it out of the obstruction, which was easy to do. The next time I put it in the same place and asked it to return to home, it flew a different route, around the tree patch, appearing to "remember" that the trees were there. It was awesome.
Watching the drone fly back home to the "start" of the shot to get ready to go again, going back to "1" as the first AD would usually say, it tried to do it directly, but got stuck in trees.
By "stuck" I just mean hovering. It never got caught in the trees, just flew into a tree pod and needed help flying out. What was impressive was the second time I told it to head back to 1, it tried a different route, retracing its steps backward.
It remembered, "Oh, the direct route is blocked by trees," so it flew a different way.
On that way back, it then managed to fly between two trees about two feet apart. A tight fit for an operator on a windy day, but exceptionally tight for a drone literally flying itself while I watched.
Tracking and the Beacon
While Keyframe is really the tool aimed at filmmakers, since this was our first time flying a Skydio of any sort, we wanted to test out the marquee feature for Skydio—tracking.
With motion tracking enabled, you get little crosses that show the object it can track. Then you select the object and it'll track it, shockingly well. This is great, but not that impressive in its simplest form; other drones can do something similar.
What's amazing is how good it is at continuing to do that when you try and crash it. I deliberately tried to run into things, to trick the Skydio into hitting a tree or a building, and it would see the building coming and elegantly fly away from it.
This all gets better if you add the Beacon, which is a device you can keep in your pocket or mount to your handlebars that offers better tracking. It also theoretically extends your signal strength, and lets you change your settings for which follow mode you have activated and set height.
For any kind of action sports shooter, these all seem like true killer features that make this platform the obvious choice for drone work.
The feature that filmmakers should be paying attention to here, and that really makes Skydio a compelling consideration for storytellers, is its new Keyframe feature.
You can fly the drone to various points and add keyframes, exactly like keyframing in an app like Resolve or Premiere. Then you can have the drone refly that same route. You can even adjust speed.
Keyframe works through a heavy reliance on the visual processing and machine learning that Skydio is building its reputation on. Sony Airpeak has a similar feature, but according to marketing, it relies only on GPS.
GPS is a factor for Skydio, of course, since GPS is a factor in most things a drone does, but it's nowhere near accurate enough to do the kinds of things you do in Keyframe. Remember, according to GPS.gov, "GPS-enabled smartphones are typically accurate to within a 4.9 m (16 ft.) radius under open sky. However, their accuracy worsens near buildings, bridges, and trees."
That's not enough accuracy to do the things this needs to do.
Skydio has cameras all around the drone body that feed into a processing engine to create a sophisticated 360° picture of the world around it. It's what allows the drone to automatically avoid obstacles in a fashion that is surprising in most of its tracking modes.
But it takes that data one step further with Keyframe, using that information to repeat shots with a high level of precision-based not just on GPS data but on visual processing data of a scene. This means it can fly closer to objects (with a few feet of a tree or light post, for instance), repeatedly.
This is huge, because the number-one thing you are always trying to do with a drone to create depth is getting close to objects. This about every crane shot you've ever seen; as they start to sweep back, a light post, flag, power line, or something else pops through frame. You want to sweep right by objects to get that sense of depth and drama.
Skydio lets you do it, over and over and over again, safely, with Keyframe.
Now, the simplest benefit of this, and the most powerful, is building really beautiful shots you can repeat take after take, worrying less about the drone piloting and more about everything else in the shot. The drone flies beautifully through most Keyframe setups (there are some minor hiccups), and you can use it as a great camera operator to help you nail amazing shots that you choreograph that might be beyond your piloting skills.
Where it gets really fun is imagining compositing.
For instance, you could do the shot five times, with the same actor in different places, and then comp them together for the actor to talk to themselves. There is a test in the press deck from Skydio that does that beautifully.
We tested it and didn't quite get the same level of accuracy; it's close, but not perfect repetition. As you see in the video below, Keyframe is great for setting up a nice drama shot and having it fly it while you watch, but doing it five times and superimposing them, it's not pixel perfect.
All of our flight days were windy (winter in New York), and I suspect you'd have better results on a less windy day. That said, it's close enough that if you wanted to pixel-tweak forever in post, you could probably get it where you need it to do composites.
We're hoping this gets refined in the future to be even more accurate, but even now, on a windless day, it seems like you can put together some great composites.
This is a huge, fascinating, amazing feature. Even with a camera on a tripod, repeating a move by hand enough to keep things lined up is basically impossible. The little, affordable Skydio being able to get it this close is astounding.
Keyframe is one of those features that is amazing and wonderful and so interesting and useful, you find yourself wanting even more. It's a little bit like hard drugs: a tiny taste just makes you ravenous for more. Or so we heard from our DARE education officer in school.
The simplest thing we would love is a Keyframe marker showing up in the recorded files. A marker for every Keyframe popping on the file that shows up in Resolve, Premiere, or Nuke, to make it easier to line up shots. Especially if you want to mix up speeds, go forward and backward, or use speed ramps, that would be huge.
Beyond that, we would love the ability to save a Keyframe "story" for use later. In the simplest setup, doing a shot at dawn, then a shot at noon, then a shot at dusk, and being able to comp them. Or doing a shot in June and a shot in December and comping them. Or, let's get crazy, shooting the same coastline year after year to watch things change. The possibilities are awesome.
But right now there isn't a "save" Keyframe button, as much as I want there to be.
That seems relatively easy to implement compared to our final hope, which is more sophisticated Keyframes. Skydio makes a unit called the Beacon, which is used for more granular tracking of subjects. Even without the Beacon, though, Skydio does some pretty sophisticated motion tracking using its neural engine. We would really love the ability to make the first or last Keyframe triggered by tracking.
Imagine a shot with a great intro view that then slides down to follow a rider. Right now to do that cool intro with Keyframe, there is no way to quickly switch to motion tracking. Linking the two together would rule, but we suspect is technically a bit challenging.
If you are now thinking, "Woah, now I can do shots like that bowling alley with no practice," think again.
Yes, shots like that bowling alley will somewhat be doable with this, but the Skdyio 2 isn't nearly as fast or as maneuverable as the racing-style drone used to get that shot. With a top speed of only 10 miles per hour, this isn't a "fast" drone. The level of skill shown in those kinds of videos is definitely beyond where Keyframe is right now.
It's weird to write an "ethics" subhead here in a film tech article, but the level of autonomy that Skydio offers means you absolutely have to bring it up. To be clear, I don't think there is anything unethical about the level of autonomy Skydio is offering, and I hope for more developments in this area from drones. But there are ethical concerns to think about.
Most drone restrictions require you to maintain a line of sight supervision over the drone and control of the drone, similar to how driving laws require you to actually keep control of the car. But the level of autonomy features you get from the Skydio 2 allows for moments of lapse. I worked extra hard to keep the controller in my hands and line of sight view of the drone at all times, so if an obstacle appeared, I could cancel an automated movement.
But it would be incredibly easy for a user to program an automated loop, set it to "go," put down the controller and run around in shot. The automated drone could then, for instance, run into a dog. Most dogs know to run from drones, but I've absolutely seen dogs run at drones, and even if you are in an "empty" park, you never know when someone is going to be walking their dog. The potential for a blade to cut a dog snoot, or worse, is definitely there.
I don't think there is a solution to this for Skydio that is easy. Maybe a solution that requires keeping a button pressed on the controller for automated moves of some sort, to at least assure that users keep their hands on the controller ready to intervene? This is particularly worrying since the platform makes it very clear it doesn't automatically avoid moving objects like cars, people, and animals.
These types of ethical conundrums are coming up around all areas of autonomy, and they are something the industry as a whole has to navigate, not just Skydio. But the point remains, you have to be just as vigilant, if not more so, when using Keyframe.
The image is perfectly acceptable. That might sound like faint praise, but it's not meant to be. It's a very nice image.
The colors are a bit too saturated, and the overall image is a bit contrasty, but all of that is useable, and a fast color grade will fix it in most lighting conditions, and "right out of the camera" I found myself pretty impressed.
The big tradeoff here for me is while it might not have the paper specs of something like the newest DJI Mavic 3, if you can use the Skydio 2+ Keyframe tool or tracking tool to get a shot that you couldn't get with another camera, the image quality doesn't need to be "as good" on paper. Image quality is about the overall quality of an image, not just resolution, and the Skydio really does let you get shots you couldn't otherwise.
The biggest frustration right now is that you don't have LOG recording, which is frustrating, since you do have 4K 60p recording, and I personally would happily trade 60p capture for 24p LOG. It also seems like there might be some chromatic aberration on some of the bright highlights in some of the skyline shots, which is less than ideal but was quite fixable in the color grade in the footage we worked with.
The drone itself doesn't fold, but it's so small when packed in its case that we didn't miss folding all that much. It fit easily in suitcases and backpacks and was snug on the front rack of my bicycle when taking it out for testing.
But I really do miss LOG, since LOG would make it more flexible.
For some of the shots we tested shooting into the sunlight on the water, LOG would've been great to help us expose to recover some of that detail in those burned-out highlights on the matter or in the shadows by the ducks. Even without that, though, we got to a grade we're pretty happy with quite quickly. LOG (and maybe someday RAW) will allow for moving a little faster, paying a little less attention to exposure, "getting the shot" in harder situations, but the image as it stands holds up quite well.
You land on the battery, which is nerve-racking. A hard landing could potentially damage the battery.
In one way this is actually brilliant; you'll likely have more than one battery, so if you do damage the "landing gear" in a hard landing, it's a replaceable part. But batteries can often get damaged without showing it, and batteries that are damaged catch fire, sometimes much later on during transit, and it just makes me nervous to land over and over on the battery. I'm sure this is a design feature that has been well thought out with a durable casting for sensitive parts, and maybe I'm overthinking it, but as someone very used to drones with top-mounted batteries, this is a weird thing to wrap my brain around.
We did all of our testing on the Skydio 2, because of availability, but the 2+ has already been announced and has a few features that we think would make it more appealing for filmmakers.
Its biggest key is extending the range of the drone from 3.5 km to 6 km, which would be a big improvement. Those numbers always refer to ideal flying conditions with little RF interference, but that rarely happens in reality, and flying in a busy RF zone like NYC I found myself noticing the limitations of the radio power in the 2 quite quickly. Even in relatively radio-quiet upstate New York, I was frequently surprised at the limited range of the original radios in the plain old 2. Beefed-up radio power for the 2+ will be appreciated.
The batteries are also revised to allow for an extra 20 minutes of flight time.
Those batteries do burn out relatively quickly, so something like the Cinema kit which comes with three batteries, an external charger (so you don't have to use the drone), and ND filters would make the most sense for filmmakers.
Would I get this?
Honestly, even with its (small) frustrations, yes. Of course, it would rule to have LOG and hopefully, that will come with a Skydio 3 in the future (maybe a Pro with Prores RAW even).
But a drone is rarely my only camera; usually, I'm getting an establishing shot for a scene or transition shots or just otherwise adding "flavor." And I can live without some of the higher-end image specs for the tremendous leaps forward you get with the autonomy functions, especially out of Keyframe.
For "gravy" and flavor and establishing shots are where the Skydio 2+ Keyframe feature really comes to life. You can program repeatable moves that are just more sophisticated, and fly closer to objects, than you might otherwise. So if you want to have the main character's car pulling into the driveway as part of the establisher, or see your actors going into a building, it's going to happen faster with Keyframe than without.
Add on top of that the potential for motion control applications, and it's just too interesting a drone not to consider it.