A doe and her fawns cross a clearing. It is a few seconds of ordinary phone video. Then, instead of playing the clip back, a developer reconstructs each moment in three dimensions and leaves it hanging where it happened. The animals dissolve into a long, translucent ribbon of poses, a tunnel of deer-shaped instants stretching across the field.
The clip went viral, and its creator, technologist Bilawal Sidhu, has since pushed the idea much further. He calls the result a “spatial memory palace.”
Quick answer: A spatial memory palace is a concept, not a downloadable app. It means placing photos and videos inside 3D models of the places where they were captured, then keeping time as a fourth dimension. Sidhu has shown it with a deer, with himself, with a conference and with a plane-crash site. It is separate from his open-source project God’s Eye View, which shows live public data (flights, ships, satellites, weather) on a 3D globe and which you can run today.
What “4D” means in this experiment
In the video, “4D” is simple arithmetic: the three dimensions of space plus time. A normal video shows one frame at a time, and each frame replaces the last. Sidhu’s version reconstructs selected frames as 3D objects and keeps all of them in the scene at once.
Animators call something similar onion skinning, where earlier frames stay faintly visible beneath the current one. Here the layers are full 3D reconstructions you can walk around.
The software that infers 3D structure from flat video is a computer-vision problem with a long history. Sidhu says he used “state-of-the-art models” but does not name them in the transcript. For that reason this article does not say which ones.
He also credits an earlier viral “cube” clip, which stacked 2D frames to visualize time, as the spark. His step was swapping the flat frames for real 3D reconstructions.
Why a deer turned into a philosophy debate
Pointed at a person instead of an animal, the same trick feels stranger. Sidhu walked through a scene and left a 3D copy of himself at every moment. Commenters reached for Kurt Vonnegut.
In Slaughterhouse-Five (1969), the alien Tralfamadorians perceive time all at once. They see humans as long, many-legged creatures, with a baby’s legs at one end and an old person’s at the other. The reconstruction renders that image almost literally: a body smeared across time.
From there the comments went in several directions:
- One viewer likened people to caterpillars burrowing through space.
- Another used the book analogy: we read one page at a time, but the whole book already exists. That is a popular way to frame the “block universe” idea from physics and philosophy, and it prompts the free-will question.
- Others brought up Chitragupta, the scribe in Hindu tradition who records every deed. Commenters said similar figures appear in Sikh tradition too.
Sidhu’s own take stays secular. If we lived in a simulation, he says, an audit trail would be a plausible feature. He also notes that “audit trail for reality” is another way to describe surveillance. That caveat is worth carrying through the rest of this piece.
From seconds to years: placing old footage in a place
A 15-second clip is a toy. The more consequential step is combining footage taken months apart.
Sidhu 3D-scanned a location in summer. He also had older videos of the same spot shot in snow. The process he describes has two parts. First, work out exactly where each video was filmed, a step called localization, by matching it against the 3D scan. Second, project the footage onto that model.
The result lets him stand in the summer scene and watch winter play out on the same ground. He describes it as navigating through space instead of scrolling through time.
The name echoes an old technique. The “memory palace,” or method of loci, is the ancient practice of memorizing things by mentally placing them in rooms of a familiar building. Sidhu’s version attaches real recordings to real coordinates rather than imagined ones.
Crowds as cameras: the TED 2026 demonstration
If one person’s footage can be pinned to a place, so can everyone’s. For this demo, Sidhu collected publicly available Creative Commons photos from TED 2026, plus his own phone shots. Because so many photos target the stage, he could infer where each was taken by anchoring on that shared subject.
He then queried the set (how many photos he appears in, he says 16) and lifted each photo into 3D at its estimated position. That put some images backstage, in the green room, and near named speakers.
These details come from the creator’s own demonstration. I could not independently verify the photo counts or the event material. The approach has an obvious dependency: it works best where thousands of people photograph the same thing. A quiet street or a private room yields almost nothing.
It also raises a privacy question. Finding every photo of a particular person is a different act from mapping where photos were taken. God’s Eye View’s repository states that it does not build features for named-person search, face recognition or tracking individuals. That policy covers that codebase. It says nothing about experiments built elsewhere.
The Miami crash reconstruction
The most serious example in the video is Prime Air Flight 7598. Here the verifiable record is worth laying out, because a 3D replay can look more complete than the evidence behind it.
On September 6, 2026, a Boeing 767-300 freighter operating for Amazon Prime Air overran a runway at Miami International Airport, struck two vehicles, and killed five people while injuring five more, according to federal investigators. The plane struck navigational equipment before hitting a van carrying seven contract cleaning workers, then another vehicle on a road outside the perimeter fence. Investigators described a runway overrun of about 1,300 feet and noted that Runway 30 has no engineered arrestor system (EMAS). The FAA does not require such systems on runways longer than 1,000 feet. article 57a723fe d8c4 4010 947c 5995e9b84ec3 +3
According to the NTSB, those who died were all in vehicles on the ground. wgcu
Sidhu says he built a 3D model of the site from drone footage and replayed public flight-tracking data inside it, then attached ground-level photos to the model. Putting a third-person view of the track beside the ground-level evidence is the most useful application in the video. I did not confirm his claim that the NTSB released the drone footage he used, nor the skid marks he describes. NTSB-published video of investigators inspecting the aircraft does exist.
Public tracking data and investigation data differ greatly in quality. Tracking sites rely largely on ADS-B, a system in which aircraft broadcast their position and speed. Flightradar24 reported the jet was moving at 112 knots when it left the usable runway. The NTSB reads something deeper. Its preliminary update said the flight data recorder held about 54 hours of data across more than 400 parameters, and the cockpit voice recorder held over two hours of good audio. wdezprimarynewssource
A replay built on tracking data shows where the aircraft went. It cannot show why. The cause had not been determined in the reporting I reviewed, and the NTSB has said it is examining everything from mechanical issues to weather and flight-crew performance. aviationbusinessnewsmichiganpublic
God’s Eye View: the part you can run
Everything above is Sidhu’s experimentation. God’s Eye View is a distinct, public project, and it is easy to confuse with the memory-palace footage.
The repository describes a browser-based “spy satellite simulator” showing live public data on a photorealistic 3D globe, including aircraft, ships, satellites, earthquakes, traffic and public cameras. It is released under the MIT license, and the README lists Sidhu and Sameh Khamis of Halfpixel as maintainers. The project’s own page says it reached #1 on GitHub Trending in August 2026, and when I checked it showed roughly 44,000 stars. githubgithub
Two details matter if you plan to use it.
“Live” is not uniform. The README says most feeds are live or regularly refreshed, but traffic is simulated along real roads using aggregate location data, and CCTV camera poses and rocket-launch trajectories are coarse estimates. It also warns the data may be delayed, incomplete, modeled, inferred or wrong, and should not be used for navigation, emergency response or other safety-critical purposes. githubgithub
The weather layer has a boundary. Its cyclone layer draws advisory positions and forecast tracks from NOAA’s National Hurricane Center and Central Pacific Hurricane Center, covering the Atlantic and eastern and central North Pacific. The video shows a storm heading toward Japan. That falls outside this layer’s documented coverage, so those visuals evidently use other sources the speaker does not identify. github
What it costs
“Open source” describes the code license. The data and services carry their own terms.
Most layers need no key or signup, including OpenSky flights at anonymous access, USGS earthquakes, CelesTrak satellites and NOAA weather. Photorealistic 3D and voice control are where charges can appear: github
- Google’s 3D tiles can come through a free Cesium ion account for personal, non-commercial use, or through a metered Google key.
- The README calls the OpenAI-powered voice feature the one that costs real money, running a few cents per active minute, with an in-app spend meter and a $5 session cap. github
- The maintainers say an official hosted version is in development at Halfpixel, so users won’t need to install anything. github
What exists today and what doesn’t
| Idea from the video | Status |
|---|---|
| Live public data on a 3D globe | Available. God’s Eye View, open source |
| 4D reconstructions of video (deer, person) | Demonstrated by the creator; no public tool identified |
| Old footage projected into a 3D scan | Demonstrated; not packaged for general use |
| Crowdsourced event reconstruction (TED) | Demonstrated; counts and details unverified |
| A personal upload service that turns your camera roll into 3D scans | Proposed. Sidhu asks viewers whether they would want one |
| A shared global “4D commons” | Aspirational |
Limits worth stating plainly
Coverage is patchy, and Sidhu says so himself: the world has pieces of a record, not a complete one. Cameras cluster at landmarks, events and places where people live, and large stretches of the planet and of the past have almost nothing.
Time is the hard part. The README makes a point worth repeating: the present is the cheap part of this field. Storing, tiling and scrubbing through what happened, at real resolution, is where the cost and computing demands climb sharply.
Then there is consent. A system that places every photo and video exactly where and when it was taken is a memory aid for the person who owns them. Pooled and searchable, it becomes a tool for locating anyone. Sidhu’s proposal for opt-in sharing, with private material staying private, addresses this in principle. It remains a proposal.
What to expect next
For now, the only piece you can run yourself is God’s Eye View. It is a live-data globe, not the memory palace. The 4D reconstructions are the creator’s experiments, shown in video. Whether a consumer service that converts a camera roll into spatial memory ever ships is an open question, and as of October 2026 nothing I found announces one.
9. FAQ
Is the spatial memory palace an app I can download?
Not as far as I found. It is Bilawal Sidhu’s term for a set of demonstrations. God’s Eye View, the related open-source project, is available on GitHub.
What does 4D mean here?
Three spatial dimensions plus time. Video frames are reconstructed in 3D and left in the scene, so you see motion as a trail rather than a sequence.
Is God’s Eye View really real-time?
Mostly live or regularly refreshed, but not entirely. Per its README, traffic is simulated, and camera poses and launch trajectories are estimates.
Is it free?
Most layers need no key. Photorealistic 3D can be free for personal, non-commercial use through Cesium ion. Voice control costs real money through OpenAI.
Can it track individuals?
The repository says it does not build features for named-person search, face recognition or tracking individuals.
Does the Miami reconstruction explain the crash?
No. It maps the path and scene. Investigators had not determined a cause in the reporting I reviewed.
