Fly it once.
Keep it forever, in 3D.
Point a drone at a place. Fly a simple pattern. Teleport turns the photos into a photoreal 3D scene you can walk through in any browser, from a single house to a whole neighborhood. This guide shows you exactly how to fly it.
What you're actually making
A 3D Gaussian splat rebuilds a place out of millions of small, soft blobs of color floating in space. Each blob knows where it is, what color it is, and how transparent it is. Blended together, they reproduce the real scene with fine detail, soft edges and natural light. You move through it freely, from any angle.
Your photos in. A walkable scene out.
You fly the drone and take ordinary photos. Teleport's cloud does the rest: it works out exactly where each photo was taken, grows the splats until they match what your camera saw, and gives you back a link that opens on a phone, a laptop, or a VR headset.
Here's the part that matters for you, the pilot: the reconstruction can only rebuild what your camera actually saw. Every rule in this guide comes from that one fact.
+ Go deeper: how the reconstruction works
Step 1, pose estimation. The pipeline matches features (corners, textures, edges) that appear in multiple photos and triangulates both the 3D structure and the exact position and rotation of every camera. This is why overlap between photos is non-negotiable: no shared features, no solve. GPS tags in your images give this step a strong head start.
Step 2, splat training. Millions of 3D Gaussians are optimized (moved, resized, recolored) until rendering them from each camera pose reproduces your photos. Surfaces seen from many angles converge crisp and stable; surfaces seen rarely (or only from one direction) stay blurry or grow "floaters."
Not generative AI. Nothing is invented. It's an optimization against your own photos, a reconstruction of what was physically there. Unlike a photogrammetry mesh, splats preserve view-dependent light (reflections, sheen) and handle soft detail like foliage gracefully, which is why they look like photographs instead of video-game geometry.
What makes or breaks a capture
Whatever you're capturing and however you fly it, a good reconstruction comes down to a few fundamentals: sharp frames, even light, and enough overlap. Get these right and the rest of this guide is just detail.
Sharp frames
Every photo has to be crisp. Blurry or badly-exposed frames don't just look bad, they drag down the whole reconstruction. A fast shutter and steady, deliberate flying keep frames sharp.
Even light
Keep the light consistent across the whole capture. The reconstruction bakes in whatever it sees, so shadows that move between passes show up as ghosting. Bright overcast is ideal.
Overlap
Neighboring photos need to share plenty of the same view, roughly 70% or more. Overlap is how the software ties photos together and resolves fine detail; too little and the scene fragments or fails to reconstruct.
Dial in your capture settings
These are the parts you actually control on the drone, and they're what keep your frames sharp and evenly exposed. Set them on the ground before you take off.
| Capture mode | Photos, not video |
| Lens | Wide angle (~24 mm equiv.) |
| Shutter | 1/500 s or faster |
| Exposure | Manual: fix ISO, aperture & shutter before takeoff |
| Shooting | Interval, every 2–3 s |
| GPS | Enabled · RTK if available |
| Dewarp | On, esp. very wide lenses |
| Overlap | ~70% between photos |
| Batteries | Enough for the flight, plus a spare |
Fly steady at around 3 to 6 m/s so interval shots stay sharp.
Capturing a house (or any one building)
The proven recipe is taking photos at 2–3 second intervals while flying three orbit loops plus cascades: circle the building three times, each loop wider and higher than the last, then fly a few vertical climbs to fill in ground-level detail. Drag the diagram to look around. The small screen shows what the drone sees.
Loop by loop
Circle the building three times, each loop higher and further out so it fills less of the frame on every pass. Loop 1 stays low and close, the house filling about ~50–75% of the frame with the camera tilted ~10–20° down. Loop 2 pulls out and up to roughly ~30–60% at ~20–30°. Loop 3 is high and wide, the house small in frame at ~10–40% with the camera at ~30–45°.
Then fly a few vertical cascades around the building, camera locked on it. Treat their height as a minimum: each cascade climbs until it at least reaches Loop 1's altitude, so it connects to the orbits. Going higher is fine; stopping short leaves those photos stranded with nothing to match against.
+ Size calculator: how high, how far?
Enter your building's width, meaning its longest side on the ground. This gives a rough estimate of the capture radius (how far out to fly) and capture height (how high) for each loop, scaled from the 24 mm-camera geometry. Treat the numbers as a ballpark to aim for, not exact values.
Building width 15 m
| Loop | Capture radius | Capture height |
|---|
Cascades fly the same radius as the low loop and climb to at least its height. Radii sit around 3–5× the building size at the high loop, which is what pulls in the surrounding context.
Capturing a neighborhood
For anything bigger than a single property, like a block, a campus or a whole district, switch strategies. Fly a grid pattern over the area like mowing a lawn, with the camera tilted forward and down at 45°, then fly the same grid again crosswise. The tilt is the point: an oblique camera captures both roofs and facades in a single pass. A straight-down (nadir) grid is great for flat mapping; here you just add the tilt so the walls get seen too.
The Drone Mission Planner
The Drone Mission Planner (Aerial 3D) turns a polygon you draw on a map into an autonomous mission. Fly the grid by hand if you prefer, or let the planner handle it. Here's the flow:
Photos taken. Now make it a place.
Drag your photos into Teleport and the cloud does the rest: no GPUs, no tuning, no software to install. A few hours later your capture is a link that opens on any phone, laptop or headset.
Common problems, and what caused them
Nearly every disappointing capture traces back to one of these. Tap any of them for the cause and the fix.
Floatersstray blobs of splat hanging in empty space+
Floaters appear where the reconstruction couldn't pin a surface down. The usual causes are too little overlap between photos, or something that moved during the capture, people, cars, branches in the wind.
Keep overlap generous (~70%+) and capture a static scene. Re-fly any thin sections so every surface is seen from several angles.
Blurry or smeared surfacessoft frames get baked straight into the scene+
A shutter that's too slow smears each frame with motion blur. Soft frames both weaken the position solve and get baked into the trained surfaces, so one slow-shutter pass can fuzz a whole facade.
Shutter ≥ 1/500 s, fly slowly and smoothly, shoot interval stills. Skim the set afterwards and delete any soft frames before uploading.
Missing or soft detail on facadesstraight-down flying never sees the walls+
Nadir (straight-down) passes only ever see roofs and the ground. Walls, storefronts and undersides are never photographed, so they reconstruct soft, or not at all.
Add an oblique, tilted pass (around 45°) so the camera sees facades as well as roofs. For one building that's the orbit loops; for an area it's the tilted crosshatch.
Quality that fades in the backgroundthe horizon and surroundings barely get photographed+
If almost every frame is a tight close-up of the subject, the background, the horizon and the wider surroundings, is barely photographed. It thins out and gets softer the further it sits from what you were focused on.
Cover whatever view matters most to you. If the background or horizon is part of the shot, fly a wider, higher pass that deliberately includes it.