About DepthSnap

What it does

Drop a JPG, PNG, MP4 or MOV into the converter. The picture is analysed to work out how far each part of it sits, and the frame is redrawn twice — once per eye — with near parts shifted further than distant ones. The result is a standard side-by-side file: an SBS PNG, or an MP4 tagged with st3d and pasp so players such as Skybox, DeoVR, 4XVR and SpaceWalker recognise the stereo layout without a manual step.

Everything runs inside the tab. Files are opened inside the tab and never uploaded anywhere; the only download is the small engine that does the analysis, cached after the first visit. That is also why there is no per-conversion cost to recoup — the compute is the visitor's machine, not a rented GPU.

Who runs it

It is an independent project by a software engineer who works on 3D vision and depth sensing (@greatYiTin). Not a company, no investors, and no team page invented to look bigger than it is.

It exists because every converter we tried either uploaded the file, watermarked the output, or put the usable resolution behind a subscription. Depth models are now small enough to run comfortably in a browser tab, so that trade-off stopped being necessary.

How long a conversion actually takes

Measured, not estimated: on a 2021 M1 Pro Mac in Chrome, 30 seconds of 1080p takes about two minutes and a two-minute clip about six — roughly three to five times the clip's length. Working out the depth is the expensive step; writing the file is not.

Headset browsers are much slower than a laptop, so for anything longer than a few seconds it is worth converting on a computer and copying the file across. The current table, with resolution, timing and peak memory figures, is on the samples page; the engine and model details are on the engine page.

What it cannot do

• It cannot recover true depth. The model guesses from a single frame, so the output is a plausible 3D rendering, not a measurement of the scene.
• Frames are processed independently. There is no temporal smoothing, so fast motion can shimmer slightly and the edges of moving objects may wobble.
• Occlusions are approximated. Hair, wires, railings and other thin structures in front of a background can tear or smear.
• It is stereo, not a light field. You get two fixed viewpoints, not motion parallax when you move your head. 360° and VR180 footage are not supported.
• Native spatial video is not an output. The files are SBS; Apple's MV-HEVC spatial format is a different container and encoder.

How the guides are written

Every guide and device page is written around what actually happens on the device it names: the exact label a player uses, where the setting lives, and what to do when a file opens flat. Playback steps come from hands-on testing on the hardware we have, and from the player's own documentation where we do not. Benchmark numbers come from our own runs on the machine listed beside them — where something could not be measured, the page says so instead of filling the gap with a guess.

Pages carry a review date and are corrected when a player update changes the steps. There are no paid placements and no sponsored links on this site.

How it is funded

Ads on the article and tool pages, served by Google AdSense. There is no subscription, no paid tier, no watermark, and nothing is locked behind an ad — the converter and the download are identical for everyone. If an ad fails to load, every feature still works.

Corrections

If a step is wrong for your device, or a number does not match what you measure, email [email protected] with the file and the player you used. Corrections are applied to the page and reflected in its review date below.