This workflow brilliantly utilizes 360-degree hardware to solve the data coverage challenges of Gaussian Splatting, making high-fidelity spatial capture more efficient than ever. It represents a significant leap in bridging the gap between physical environments and professional-grade digital 3D assets.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
This 360° Drone Makes INSANE 3D Scans | Antigravity A1 & Gaussian Splatting WorkflowAdded:
Did you know that you can turn any video into a 3D scene that you can actually move through? Literally change the camera angle after filming and create shots that would be impossible to capture in real life. All thanks to something called Gaussian splatting. And I've got so obsessed with it to the point where I started experimenting with bigger and bigger scenes. Goian splatting works by analyzing hundreds of frames from your footage and figuring out where everything sits in 3D space and it represents the scene as millions of tiny colored blobs. Together they form something that looks and feels like the real place. You can use any device to make this happen. And the core process is pretty much the same across all of them, but the S tier one of them all in my opinion is using a 360 drone.
I will explain why. And I'm also going to show you how I use the anti-gravity A1 to scan any location and turn it into a 3D world that can actually be shared and edited in tools like Blender and After Effects. Oh, and if you actually try this yourself, there is a challenge running right now where you can win real cash prizes for your scans. We'll get to that. But first, let's look at how to scan buildings using a drone.
To show you the process, I chose this beautiful Chinese temple. I scanned this place before using a regular drone, specifically the DJI Mini3 Pro, and the results were decent. But I was very curious to see how much different the scanning experience is going to be if I came back this time with the anti-gravity A1, which is a 360 drone co-engineered with Insta 360. Now, I've never flown with goggles before this.
I've never touched an FPV drone. I did one short test flight the day before to understand the controls and then came here for the scan. First few seconds in the gogles were such a new experience.
>> Looks beautiful, man.
>> It's so immersive. You almost forget that you're flying a drone. It genuinely feels like you're actually moving through the space yourself. You simply point the controller where you want to go and the drone follows. And while you're flying, you can actually look around freely. Wherever you turn your head, that's what you see in the goggles. I'm not going to lie, I was a little overwhelmed at first trying to fly and focus on the subject at the same time until I realized that I don't have to think about framing that much because the camera is capturing everything around it. You just move through the location and figure out the framing in post and the video remains extremely stable. Okay, but how do you actually scan a place like this? The core principle here is coverage. You need footage of your subject from every angle. And for a building, that means orbiting around it while filming.
Luckily, the A1 has a built-in feature that handles this automatically.
>> So, it's doing it all on its own. I don't have to do anything now.
>> While hovering, you simply pull up the menu, tap Sky Genie, select orbit, set the building as your target, then start orbiting, and the drone flies the entire circle for you, recording everything in 8K 360 while you just watch. That was so clean, dude.
And because it's 360, every orbit gives you far more data than a regular drone would. You're not just framing what's in front of you. You get a complete sphere of information from every position along the circle. The one area where a regular drone still has an edge is image quality. The A1 shoots 8K, but that's spread across a full 360 sphere, but we will see how much difference that practically makes in the reconstruction stage. For most of my scans, I do three orbits at different heights. A medium angle parallel to the ground, a higher or near top-down pass for the roof, and ideally a low angle too to capture hidden geometry. Here, though, between the trees, the power lines, and the fact that this is an active temple, I kept my distance and did three higher and safer passes. It's enough given the limitations, but if you can fly lower around your subject, make sure you do it. One thing worth noting, the orbit feature creates a separate recording for each pass. If you're comfortable enough to fly manually and change height mid orbit in one continuous take, that could actually work in your favor later, and I will explain why when we get to the processing stage. Now, orbiting with a traditional camera drone is possible, but genuinely stressful. The camera has to face the subject the whole time, which means the drone ends up flying sideways. And if your drone doesn't have side sensors, you're basically blind in the direction of travel. The A1 now has omnidirectional obstacle avoidance, which means you can fly with much more confidence. One thing I do wish the A1 had, though, is the ability to control the orbit speed. It's not a dealbreaker, but flying slower means more frames and less motion blur, which also means better reconstruction data. All right, so now the scan is done. Three passes, three separate recordings. We can now bring that over to the computer and start processing. But first, let's take a look at the footage itself. Here I'm using the anti-gravity studio where you can navigate through the footage in different viewing modes, get a feel for what was captured, and even create keyframed animations directly from your recordings. But this is not where the reconstruction into a Gaussian splat happens. For that, there are many options. The first one I want to show you is called Splatica. One of the easiest tools that I've come across. It has specific support for the anti-gravity A1. You can select it as your input device and after processing you will have a Gaussian splat that you can view and share online. Splata is a great option for reconstructing scans recorded in one single take. This is exactly why I mentioned that doing one continuous flight and changing heights mid orbit could actually be a valid approach. Okay, now before I show you the tool I use to process multiple videos at once, Intgravity and Insta3 are running an open challenge right now called Project Eternal. It's about using 360 imaging to archive important places around the world and you can submit your own scan for a chance to win up to $3,500. If you've ever wanted a reason to actually go out and scan something, this is it. Link in the description. Go check it out. Now, to process my footage, I'm going to be using MIMAP, and I chose it because it accepts multiple video inputs at once. The free plan lets you process up to 500 images per reconstruction, which, as you will see, is actually enough to get solid results. To get started, create a new project, give it a name, and then over to the left, you will see the video option. This is where you import your 360 footage. I'm bringing in the three orbit passes I recorded at the temple.
Next, you set the time interval for each video. I set mine to 0.5, which means midmap will extract images every half second. And because the A1 has two fisheye lenses, one on top, one on the bottom, each extraction gives you two frames, one from each lens. And with my three videos at 0.5 interval, that comes out to around 340 images, still under the 500 image limit. If you were flying slowly, you can get away with a higher interval. But if you're moving quickly, you will want a lower interval so you're not getting big gaps between images.
After extraction, you will see this setting menu where you can make some changes. Under textured model, I enabled FBX and OBJ. These are the formats that matter if you want to bring the model into Blender or a game engine later.
Obviously, you need to enable Gaussian splatting. If you had people walking through during the scan, make sure you enable people removal. After that, you can go ahead and hit start reconstruction. The processing time varies a lot depending on your hardware and your image count. I will put my specs and the time it took on screen here. And I'm going to jump to reconstruction I already have finished using the same input. You can toggle between 3D mesh mode and Gaussian splat mode down in the bottom left corner. And honestly, there's something about the Gaussian splat view that I find more visually appealing. It just looks a bit richer, more detailed, and I think it's the way the splats handle light and color. For this scan, I focused mostly on the main section of the temple because it was the easiest part to orbit, and I could get closer to it than the rest of the structure. And you can really see that in the result. The weaker areas are the ones I couldn't reach from a low angle, like you can see here. And that's mainly a coverage limitation. If I had been able to fly lower and closer, those areas would look just as good as the rest. Now, I also ran the same input through with a much higher image count, around 1,700 images.
And the difference is honestly hard to notice, which tells you that what matters most for reconstruction quality isn't how many frames you extract, it's how many angles you covered during the flight. Which brings me to this reconstruction. same location, but this time I fed Mip Map everything I had. 11 videos in total. Not just the three orbit passes, but wider angles as well and a mix of automatic orbits and freehand flying all in one project. The drone was so fun to fly, I couldn't help but record so many videos. Over 4,400 images processed. And this is what full coverage looks like. The entire structure, all of it reconstructed. You can see the dragons on the roof ridges, the lanterns, the small details, the parallax. Even with the A1's 360 image quality being lower than a normal drone, the result looks really good, even better than what I got using DJI. It takes longer to process this many videos and the output files are much heavier, but if you have the footage and the hardware, this is the way to get the best results possible. I also scanned this massive stadium and again I couldn't get much of the low angle this time because as soon as I lowered the drone behind the stadium the signal starts getting sketchy. I can't say for sure but the metal structure likely caused interference. So I had to stay higher than I wanted to. But what I could capture from above was genuinely stunning. I managed to include the pitch and the stands which gave me this incredible sense of depth when you move through this plat. Flying around the place was one of the best drone experiences I've ever had, and I came back with scan footage and some cinematic shots at the same time. For this one, I ditched the orbit feature entirely and just flew manually underneath, above, through, around, in whatever direction felt like it was giving me good coverage. And it works.
As long as you're capturing enough angles and overlap, the software figures it out. Once reconstruction is done, MIP Map saves everything to an output folder. Mine is in my desktop, but you can change the save location in the settings before you start. Inside the output folder, you will find subfolders for each format. The GS folder is where your Gaussian splat files are exported.
On the 3D side, the FBX output from MIP map imports straight into Blender. You get a textured mesh of the entire scene that you can manipulate, light, and render however you want. I won't go deep into this here, but the possibilities are pretty wide once it's inside a proper 3D environment. And you can even bring the ply file into After Effects using the Gaussian splatting plug-in.
You can create camera movements through the scene, do model reveals, even combine multiple scans together. I actually made a full separate tutorial on this showing exactly how to set it up and create those kind of animations. If you want to go deeper, make sure you check it out. Gaussian splatting is moving fast and getting so much more accessible. The A1 makes that even easier because you're not fighting the drone or thinking about framing. You're just exploring while capturing real data. Big thanks to Anti-Gravity for sponsoring this video. And if you're planning to get the A1, make sure you check the description because I've got an exclusive discount for you. If you end up scanning locations, I would genuinely love to see what you come up with. So, make sure you tag me and Intravity on Instagram. Stay creative and I'll see you in the next video.
Peace.
Related Videos
Beyond Robotics | European Rover Challenge 2026
beyondrobotics
189 views•2026-06-01
Beatbot Sora70: JetPulse Technology and AI obstacle avoidance and navigation!
DroidModderX
26K views•2026-06-02
Tesla FSD 14.3.3 Hits Phoenix Streets - FIRST LOOK
anthonystesla
114 views•2026-05-29
Elon Musk Just Revealed Fremont Line for Optimus Gen 3 Mass Production
TheAINexusOfficial
180 views•2026-05-30
人機一体「零式人機 ver.2」 子ども企画【おもしろ発見!モビリティー】 #乗り物 #automobile #robot #shorts
KyodoNews
1K views•2026-05-28
China’s New Luna AI Robot Looks Shockingly Human...
NextGenHumanoids
850 views•2026-05-28
Reachy Mini: the $300 open source robot you can actually hack — Andres Marafioti, Hugging Face
aiDotEngineer
662 views•2026-05-29
柔軟指×AI画像処理食品の仕分け作業システム!#柔軟指 #ロボット #自動化 #製造業をもっと盛り上げたい
KiQ_Robotics_Corp.
113 views•2026-05-28











