DepthWeaver turns LiDAR scans, on-device AI depth, 3D models, and objects you scan into autostereograms you can stare into — woven on your iPhone or Mac with a Metal compute kernel, never sent anywhere else.
The main screen is now the image itself, edge to edge. Controls live in a frosted-glass drawer that expands out of a floating bar — Depth, Pattern, Tune — and slides away when you just want to look. Pinch to zoom, tap to hide.
One instruction at a time — hold it at arm's length, look through the screen, let the dots double up — with a progress arc around a live stereogram. Each round you clear gets a little harder. Offered on first launch, and always behind the “?” button.
One tap for a good-looking result, then fine-tune. The sliders speak plain words now — Depth, Spread, Smoothness, with the technical term as a hint — and DPI, depth ranges and denoising move behind Advanced.
The depth source screen shows a live 3D point cloud of the depth you just captured, so you know what you are working with before you weave it. On iPhone and Mac both.
Also in 1.2: AI depth estimation no longer blocks the interface, so the app stays responsive while it thinks; the depth source screen is re-ranked around what people actually reach for first; and the 3D model screen, the full-screen result and the depth adjustment controls have all been rebuilt to match — plus a pile of fixes. The Mac app keeps its document-style inspector layout and picks up the 3D depth preview.
On iPhone Pro, the LiDAR scanner streams a real-time depth field. Tap to freeze a frame and feed it straight into the generator. (iPhone only — there is no LiDAR on Mac.)
Depth Anything V2 runs on the Neural Engine on iPhone, on the GPU on Mac. Pick a picture, get a depth map. Nothing uploads — the model is bundled in the app.
Drop a 3D file in, orbit the camera, capture the view-space depth. Bundled samples ship with the app to play with right away. Available on both iPhone and Mac.
Apple's guided photogrammetry, embedded in the app. Walk around a small object, let RealityKit reconstruct a USDZ on device, then weave its depth. Saved to your private "My captures" library — rename, reuse, delete. (iPhone Pro only.)
Six procedural pattern generators — random dot, stars, Perlin, Worley, Voronoi, reaction-diffusion — plus a curated set of bundled tiles, plus any photo from your library as a custom tile.
DPI, depth strength, separation, oversampling. Remap input and output ranges to push subjects forward or back. Denoise quantized 8-bit imports for clean, banding-free output. Invert depth in one tap. The preview redraws as you slide.
Under the hood: W. A. Steer's extension of the Thimbleby–Inglis–Witten autostereogram algorithm — link-based hidden-surface refinement, centre-outwards bitmapped patterns, OKLab gap fill and oversampling. The whole pipeline runs as a Metal compute kernel, one thread per output row, with a parallel CPU fallback when Metal is unavailable. Powered by ARKit, Object Capture, Core ML, SceneKit, Metal, and Core Image. Output is upscaled so the minimum dimension is at least 960 px — print-ready or share-ready.
One SwiftUI codebase, two first-class apps. iPhone is canvas-first —
a full-bleed stereogram with a glass tool drawer over it; Mac uses a
document-style window
with a collapsible inspector, standard menu commands (Open ⌘O, Save
⌘S, Copy ⌘⇧C, Toggle Inspector ⌘⌥I), and a .fileExporter
PNG save. LiDAR live capture and Object Capture scans stay on iPhone
where the hardware lives; everything else — AI depth, 3D models,
presets, denoising, the Metal renderer — works identically on both.
Everything runs on device. No account, no upload, no analytics, no third-party SDKs, no ads. Your photos and 3D scans never leave your iPhone or Mac. Read the privacy policy.