Real World — Theo Real World
Turn photos of a real person, object or place into a 3D capture of the thing in its surroundings — the table it sits on, the room around it — that you can spin in the chat. Walk around the subject, take photos (or a short video), and Theo Real World works out where every shot was taken and rebuilds what the camera saw: a quick capture in about a minute, a full one a few minutes later, both on the same card. Nothing is invented.
How It Works
1. Open Real World. Pick Real World from the mode selector in the composer, or tap Real World in the Theo Chat sidebar. The mode sticks to the conversation.
2. Walk around it. Move the camera slowly all the way around the subject and take photos as you go — overlap each shot with the last. Every side you photograph is a side the capture will have; a side nobody photographed stays empty. A slow 20-second walk-around video is the best input of all.
3. Drop your photos in — or the video. Add the photos all at once (up to sixty), or tap Add a video and pick one walk-around clip (MP4, MOV or WebM, up to 100 MB) — you can also drop it straight onto the chat. Theo reads what the photos are of — a person, an object, or a space — and shows its guess at each photo's angle on the board as a hint. You never have to label anything: the angles that matter are worked out from the pictures themselves when you send. With a video attached, the capture is built from the video's frames and any photos beside it are set aside — one source per capture.
4. Send it. Theo solves the camera positions and builds the capture. The quick capture lands on the card in about a minute and you can spin it right away; the full capture replaces it a few minutes later. You can keep chatting, close the tab, and come back.
Coverage Is Quality
- Move the camera, not the subject. Turning an object in front of a still camera reads as one angle, not many — the background doesn't move with it, so the shots all come from the same place. You walk; the thing stays put.
- Overlap every shot. Each photo should share a good part of its view with the last one. That overlap is how the camera positions get solved.
- Read the coverage line. Under the capture the card says how many shots went in and which sides were seen — "captured from 9 shots · seen from the front only", or "the back wasn't seen". That is the side to shoot next time.
- Faint camera markers in the viewer show where each photo was taken from, so a thin side and a missing angle line up on screen.
- Fewer is fine to start. A handful of photos already produces a capture of the side they covered; all the way around, or a video, produces the whole thing.
Person, Object, or Space
Person
Someone standing still while you walk around them. Captured where they stand.
Object
A product, a prop, a model, a chair — captured in its place, on the surface it sits on.
Space
A room, a shopfront, a garden. Walk through it while filming and get a capture you can move around in.
Theo recognises the subject from the photos and switches the picker for you when it's sure (it says so, with an Undo) — and never overrides a subject you picked yourself. The subject labels the capture and helps frame it; it never changes what gets rebuilt.
The Quick Capture and the Full Capture
- The quick capture lands first — a lighter version, sharp enough to see what you have and which sides are missing. Spin it while the full one trains.
- The full capture is the same scene refined for several more minutes: finer detail, cleaner surfaces, truer colour as you move around it. It replaces the quick one on the card when it finishes.
- Both are faithful. Neither tier invents anything — the far side of something nobody photographed stays empty in both, and the card says so.
Downloads: the capture saves as a compact .spz splat file that any splat viewer opens.
Looking Around It
- On the card, drag to orbit and scroll to zoom. The capture pivots on the subject and starts from where the first photo was taken.
- Full screen, the same controls plus a Walk button. In walk mode you drag to look, use W A S D or the arrow keys to move, Q / E to rise and sink, and hold Shift to hurry — or right-drag, scroll, or use two fingers on a phone. The ground stays under you; a room is walked, not tumbled. Tap Walk again (or the turntable) to go back to orbiting exactly where you are.
- View in AR (beta) appears on phones and headsets whose browser supports it. It sets a miniature of the capture — about a metre across — on the floor an arm's length in front of you, so you can walk around it in your own room. A capture has no real-world scale of its own, which is why it is shown as a model rather than life-size. Escape, or the same button, brings you back.
Shooting Tips
- A person should hold still, arms slightly away from the body, feet a little apart. You walk; they don't.
- An object should stay where it is. Leave it on the table and circle it — fill most of the frame and keep the whole thing in every shot.
- A space is best filmed while walking slowly through it with the camera level, not turned from one spot — the walls need to appear in several shots from different places.
- Keep the light the same for every shot and skip the flash. Matte things capture best; glossy, mirrored, transparent or very thin things come out soft.
- Video beats photos. A slow, steady 20-second walk-around gives hundreds of overlapping frames — the single biggest thing you can do for quality.
The studio lists the tips for whichever subject you picked, right under the board.
What to Expect
The quick capture usually lands in about a minute (a little longer when the scanner is cold); the full capture takes a few minutes more, longer for a video with many frames. The card shows the step it's on and never claims something is finished before it is. If the full capture can't finish, the quick capture stays on the card and a note under the viewer says so.
Credits are charged once per scan — a quick capture on its own is cheap, and quick + full together cost about as much as a Standard 3D render. A failed full capture is never charged twice.
To capture again, re-send the photos (with more of whatever side the coverage line said was missing). A capture needs the pictures themselves, so the card doesn't offer a Regenerate button the way a 3D render does.
What It Doesn't Do
Theo Real World captures things that exist. It won't invent a model from a description or dress up a single reference picture — that's 3D mode, which builds an object from a prompt or from a few view-tagged photos. If you have the real thing in front of you, capture it here; if you have an idea, build it there.
It captures one subject at a time, doesn't rig or animate the result, produces a splat rather than a printable mesh, and can't fill in a side you never photographed — if the back looks thin, the coverage line will already have told you why.
Related Articles
3D Mode
Turn a photo or a prompt into a textured 3D model, spin it in the browser, download it as GLB/OBJ/STL/FBX/USDZ, and split it into interlocking parts for 3D printing. To scan a real person, object, or space from photos, see Real World.
Theo AI Overview
Overview of Theo AI — 11 chat modes, creation tools, generative UI, memory, and the sidebar mascot.
AI Chat
Conversational AI that creates flowcharts, whiteboards, notes, and presentations from natural language.
Video Mode
Generate videos from chat prompts — choose styles, durations, and aspect ratios.