SHARP, an approach to photorealistic view synthesis from a single image

Existing and Likely Product Uses

• Widely assumed to be behind Apple’s Cinematic/portrait-style effects and “Spatial Scene” parallax wallpapers and Photos features.
• Seen as an aesthetic differentiator (lock screens, album covers, Vision Pro spatial photos) more than a core “productivity” feature today.

What SHARP Technically Does

• From a single 2D photo, it infers a 3D point-cloud / gaussian-splat representation (with camera parameters) and renders novel nearby views.
• Enables parallax, dynamic IPD for stereo/VR, and slight camera motions that preserve texture and lighting (“photorealistic” vs flat depth maps).
• Distinct from NeRF-style opaque latent fields; here the intermediate 3D structure is explicit and exportable (.ply splats).

Perceived Quality vs Prior Methods

• Many find the results visually impressive and sharp, especially compared to earlier parallax tricks.
• Others note artifacts: broken water reflections, warped skies, ghosting around edges, “2D cutout” feel for people, and “nightmare fuel” failure cases.
• Comparisons with TMPI and other methods are mixed: SHARP often wins on realism and depth consistency, but not universally.

Implementation and Accessibility

• Official repo requires CUDA for trajectory/video rendering, prompting frustration from Apple-silicon users.
• Multiple comments confirm the core model runs on CPU/MPS; gaussian splats can be exported and viewed in WebGL/other viewers.
• Community forks demonstrate it running on Apple Silicon, though some early demos look rough.

Imagined Applications

• Entertainment: phone photo enhancement, VR stereo pairs, “Ken Burns++” history docs, converting old photos into subtle 3D shots.
• Simulation/CAD: interest in turning photos into usable 3D geometry to speed up asset creation, though some doubt it’s accurate enough for robotics/physics.
• Archival and stereo collections: curiosity about feeding stereo pairs or video sequences to improve quality.

Concerns and Skepticism

• Some question the economic value of ever-better visual fakery vs more reliable reasoning/knowledge AI.
• A dystopian thread worries about hyper-immersive media leading to social withdrawal and “wireheading,” though others argue such existential fears are recurring and not definitive.