Drop one photo and a neural network in your browser works out what is near and what is far, lifts the subject off the background, fills in what was hidden behind them and moves a camera through the scene. The result is a short video that looks filmed, made from a picture that was never meant to move. Free, no watermark, nothing uploaded.
✨ Make a 3D Photo →Free · No watermark · No sign-up · Your photo stays on your device
First, a depth model estimates how far every part of the picture is from the camera. Where the depth jumps sharply — the edge of a person against a wall, a head against the sky — the image is split into a near layer and a background. The part of the background that the subject was covering does not exist in the photo, so a second model paints it in, matching the texture around it. Then the scene is rendered from a camera that moves: the near layer and the background shift by different amounts, exactly as they would if the shot had been filmed while walking past.
All of it runs in the browser. On a computer with a recent graphics card the depth takes well under a second; without one it takes a few seconds per photo, and a phone needs longer. The models — about 50 MB together — are downloaded the first time and then reused.
The usual “3D photo” trick stretches a single image over a depth map. As soon as the camera moves, the pixels along the subject’s edge smear into the gap that opens behind it, and the eye reads it as fake. Splitting the photo into two layers and filling in the background removes that smear: when the camera slides, there is real background behind the person, not a rubber sheet. That is the difference between an effect and a camera move.
Orbit traces a slow figure around the subject — the classic look. Sway rocks from side to side. Push in slowly moves closer, with the near layer approaching faster than the background. Circle keeps an even round path. Then Volume sets how strong the separation is, Speed sets the tempo of the move, and Punch decides how hard the camera kicks on the beat. By default the punch is off: the move stays calm, and you add energy only if the track asks for it.
With a song added, the camera can jolt on the kick and the bass. With the beat switch in the top bar turned off, nothing pulses at all — the photo just drifts in depth, which is the look that suits a ballad, a portrait or a cover. Without any song, a video without music gives you exactly that: set a length from 5 seconds to 5 minutes and export a file with no audio track — a Spotify Canvas, a status, a living wallpaper.
Portraits and pictures with one clear subject and some distance to the background give the strongest result: the model sees a clean jump in depth and the layers separate neatly. Wide landscapes work too, with the foreground sliding over the horizon. Flat images — a poster, a screenshot of text, a photo of a wall — have no depth to find, and the move stays subtle. Glass, mirrors and very fine hair against a busy background can confuse the edge; sharp, well-lit photos with room around the subject avoid that.
3D PHOTO is a card of its own, but the depth it computes is reused elsewhere. Any photo frame can get its own 3D camera under another effect — VHS, glitch, film — and a slideshow of up to 30 photos can give every frame its own move. SMOKE uses the same depth map to find the person and put the light behind their head.
A release with no music video. A press shot or the artist photo from the cover, turned into a slow camera move, carries a YouTube upload or a Reels post far better than a static image — and it takes minutes, not a shoot. A Spotify Canvas. Eight silent seconds of depth behind the track, made from the cover photo. The first seconds of a Short or a TikTok. Motion in the very first frame is what stops the scroll, and a 3D move gives it without any effect that looks like an effect. A lyric video background. The camera drifts through the photo while the words sit on top. And outside music — a travel photo with the foreground sliding over the horizon, a portrait for a story.
Leave room around the subject: the camera moves, and a head touching the frame edge will be cut. Use Push in for a close portrait and Orbit for a wider shot where the background has detail to reveal. If the background is busy — a crowd, leaves, a pattern — lower Volume so the layers part gently. Keep Punch at zero for anything slow; turn it up only when the track hits hard and you want the camera to hit with it. And if the export is for a feed, check the 9:16 crop first: the automatic crop option puts the subject in the middle for you.
| Photo in | JPG, PNG, WebP · one photo, or up to 30 in a slideshow |
| Camera | Orbit, Sway, Push in, Circle · Volume, Speed, Punch |
| Audio | your song (MP3, WAV, FLAC, M4A, OGG) or no audio, 5 s to 5 min |
| Video out | MP4 (H.264) · 720p, 1080p or 4K · 9:16, 1:1, 16:9 |
| Models | about 50 MB, downloaded once · computed in the browser |