Apple's SHARP turns a single photo into 3D in under a second

Apple's SHARP turns single 2D photos into detailed 3D scenes in under a second

Apple's new neural system, SHARP, takes a single flat photo and produces a high-fidelity 3D scene in less than a second on standard hardware. It builds 3D Gaussian Splatting (3DGS) representations with speed and accuracy, landing a reported 25-34% boost in visual quality over prior models. The headline: near-instant 2D-to-3D with measurable gains, and it's open-source.

How it works (without the fluff)

SHARP uses 3D Gaussian Splatting-think of it as a smart point cloud where each point carries color and density modeled by Gaussian functions. Instead of just guessing depth, the network learns object boundaries, textures, and spatial hierarchies to reconstruct full scene structure from a single image.

The output is saved in a 3DGS format, which plays well with modern renderers and visualization tools. If you want a quick primer on the technique behind it, see the original 3DGS research and demos here: 3D Gaussian Splatting.

Why this matters for devs and IT

• Single-image scene reconstruction cuts capture costs compared to photogrammetry that needs dozens of angles.
• It runs on standard GPUs in under a second, which means practical integration in apps, pipelines, and services.
• Outputs a portable scene format (3DGS), making it easier to slot into Unity/Unreal workflows or custom viewers.
• On Apple hardware, Neural Engine acceleration plus LiDAR can form hybrid workflows: sensor depth + AI reconstruction.

Apple's ecosystem play

SHARP lines up with Apple's push into spatial computing. A related feature called Spatial Scenes is slated to debut with iOS 26, tapping Apple's Neural Engine to generate 3D from 2D photos directly on-device.

Apple has also published learning resources showing how to swap sample images with your own to spin up interactive 3D objects for classes, design, or creative projects. With LiDAR-equipped iPhones and iPads, teams can blend real depth capture with AI inference for cleaner geometry and faster results.

Real uses you can ship soon

• E-commerce: turn catalog photos into 3D previews and let shoppers place items in their rooms without extra scans.
• AR/VR production: bootstrap Vision Pro experiences by generating scenes and props from reference photos.
• Architecture: instant 3D previews from site photos for concept reviews and client walkthroughs.
• Education: convert museum archives or medical diagrams into explorable 3D content for more active learning.

Competitive context

Other teams are moving here too. Tools like PartCrafter (from Peking University, ByteDance, and CMU) focus on object modeling, while SHARP targets scene reconstruction-better aligned for VR, visualization, and storytelling than manufacturing or 3D printing use cases.

For developers: practical first steps

• Test the SHARP repo and sample scenes on a mid-range GPU; benchmark inference time and VRAM usage on your stack.
• Define your pipeline: single image in → 3DGS out → viewer/engine import (Unity/Unreal or a lightweight WebGL viewer).
• Measure quality: compare PSNR/SSIM/LPIPS against your current pipeline to validate the 25-34% improvement claims.
• Prototype hybrid capture: augment a single image with optional LiDAR depth when available to stabilize fine geometry.
• Plan privacy: if you're handling user photos, aim for on-device inference where feasible to reduce data movement.

PRO TIP: Open-source releases from Apple often foreshadow new APIs. Track SHARP closely-expect touchpoints with future iOS updates and frameworks.

Try iOS 26 and iPadOS 26 betas

iOS 26 and iPadOS 26 ship with a stack of spatial features, and you can test them early. Check your device against the supported models list, then enroll via Apple's Beta Software Program-no paid developer account needed: beta.apple.com.

What this signals for visual computing

By open-sourcing SHARP, Apple is inviting the broader community to iterate on the approach and bring it to more platforms. Sub-second performance on standard GPUs opens use cases like real-time scene generation during calls and instant architectural previews from a single photo.

As features like Spatial Photos, Spatial Scenes, and the Vision Pro ecosystem mature, 3D won't feel special-it'll feel default. SHARP is a clear step toward that future: fast, accessible 3D from everyday images, integrated into tools people already use.

Level up your spatial and AI skills with curated training and tools: Latest AI courses.