Image to 3D depth effect

How a flat photo yields depth: monocular estimation in plain words

You can look at any ordinary photo and instantly tell what is close and what is far — even with one eye shut. You are using learned cues: near things sit lower in the frame and overlap distant ones, familiar objects have expected sizes, textures compress with distance, and far-off scenery goes hazy. The AI model on this page has learned the same cues from millions of photographs and applies them to yours: for every pixel it estimates how near or far that point is, producing a depth map — effectively a grayscale relief of the scene. That map then drives the 3D effect: pixels judged nearer are shifted further apart between the left-eye and right-eye views, which is exactly the difference your brain reads as depth.

Relative depth — what that means for your result

The model tells you what is nearer and what is farther, never how many metres away anything is. In practice that has three consequences. You cannot measure with it — the output is an illusion for eyes, not data for engineering. The depth ordering is occasionally wrong in ways human vision would not be: mirrors, water, glass, featureless skies and photos of photos can confuse it, because reflection and transparency break the cues it learned from. And along sharp foreground edges you may notice a thin fringe where shifted pixels reveal areas the camera never actually saw — the effect fills them in plausibly, but a close look shows the trick. Photos with one clear subject at a distinctly different distance from the background give the most convincing depth of all.

Anaglyph or side-by-side: two outputs for two audiences

The anaglyph output is the one to share: a single normal-sized image that anyone with a pair of cheap red/cyan cardboard glasses can view — on any screen, and even printed on paper. The price is colour accuracy, since the effect works by sending different colour channels to each eye. The side-by-side output keeps full natural colour but asks more of the viewer: it is two complete views next to each other — the file is exactly twice the width of the source — meant for a VR headset, a stereoscope or the cross-eyed free-viewing technique. A fair rule of thumb: anaglyph for sharing with an unknown audience, side-by-side for your own viewer. On strength, start subtle — a modest pixel shift reads as clean depth, while a large one starts to tear the edges apart.

The one-time download, and what stays on your device

Everything here runs inside your browser. The depth model is downloaded from our own site on first use — about 50 MB, cached by your browser afterwards, so only the first conversion carries the wait — and your photo never leaves your device at any point. To keep the estimation quick, the working image is scaled down to at most 768 pixels on its longest side, and the 3D image is built at that size — fine for screens and sharing, which is what a depth effect is for. The free tier covers five photos per day on this tool, up to 10 MB each. A last practical tip: the side-by-side output being double width makes for a hefty file, so run it through our free Image compressor before posting — a much smaller file, with the stereo effect fully intact.

This is a depth effect, not a 3D scan

The AI model (Depth Anything V2) estimates <em>relative</em> depth — what's nearer or farther within the photo — not real-world measurements. That depth map shifts pixels sideways by an amount proportional to how close they are, which is exactly what creates a stereoscopic illusion when viewed with red/cyan glasses or by cross-eyed viewing. Think of it as a stylized 3D effect rather than a precise 3D reconstruction.