Elliott Tate has released Spatial Slideshow 0.9.0, a free MIT licensed Mac app that turns Photos albums into moving 3D scenes.
Spatial Slideshow runs the Reframe models that Photos installs on the Mac, loaded in place from the SpatialPhotosRelive asset directory, and renders the result through Apple's own Gaussian renderer. According to the project's research notes, the predictor returns the scene as Gaussian splat buffers, which the app passes to the GSAsset, GSSorter and GSRenderer classes in the private CoreRE3DGSFoundation framework on macOS 27.
Apple introduced spatial scenes in June 2025 and described them as the product of "a new generative AI algorithm and computational depth." Tate's findings are third party reverse engineering, but they indicate that the Reframe model Photos installs under its SpatialPhotosRelive assets outputs Gaussians. Apple renders splats publicly through RealityKit 27 and its GaussianSplatResource, while Spatial Slideshow calls the private framework directly.
The scene step color manages each photo, runs the predictor and saves the buffers. A camera then moves through the splats, so foreground and background separate with depth, and a foreground aware depth estimate keeps distant sky or fog from producing excessive travel.
Version 0.9.0 adds a real time 3D mode that animates cached scenes on the GPU at a 60 fps target and skips MP4 encoding. Scenes persist in a 4 GiB least recently used cache. A paused photo can be explored with the mouse, trackpad or WASD keys, motion strength reaches 4x with orbiting paths, and a native macOS screen saver plays prepared slideshows. Output reaches 3840 pixels on the long edge, and optional edge expansion uses Apple Fast Clean Up, FLUX.2 Klein through MLX, or FLUX through Draw Things.
Spatial Slideshow requires Apple Silicon, macOS 27 and installed Photos Reframe assets and bundles no model weights. Tate states that the private frameworks and models are verified only on his development machine, and that an OS update may require source changes. Release builds are Developer ID signed and notarized.
Each photo whose scene is still cached can be replayed along a new camera path or explored by hand without running inference again.
Spatial Slideshow 0.9.0 is available now on GitHub.
