WOM -- World of Memories: SOTA Agentic 3D Reconstruction with Measurable Fidelity

# WOM -- World of Memories: SOTA Agentic 3D Reconstruction with Measurable Fidelity https://wom.quest try it out! **What if you could reconstruct an entire world from a single photograph—with measurable precision?** Image-to-3D is no longer new. The frontier is fidelity: can a generated world preserve what a photograph actually shows, remain coherent when explored, and improve as the models improve? WOM turns photographs into explorable 3D worlds. Move beyond the original camera view, navigate reconstructed scenes, capture new perspectives, and create new images and eventually films from old memories. **Generate.** Meta Muse Spark and Muse Wonder and SAM3 orchestrate our AI3D infrastructure across AWS, GCP, Hugging Face, Replicate, and fal, coordinating specialized image-to-mesh and image-to-splat models including World Labs Marble Beta v2 and SOTA image2gassian SperidLabs IDOL. Spark Radix WebGPU powers explorable Gaussian-splat worlds populated with interactive splat objects. **Inspect and Improve — WOMAgents.** Autonomous agents navigate reconstructed environments, investigate novel viewpoints, detect visual and geometric aberrations, and attempt targeted corrections. A world should hold together beyond its original camera angle. **Measure — WOMetrics (World Observation Metrics).** Our open, model-agnostic framework evaluates visual fidelity, geometry, object placement, completeness, and multi-view consistency. We aim to turn the noise surrounding frontier AI3D research into comparable, reproducible evidence. **Compete — WOMPar & WOMgolf.** WOMPar is our best reconstruction of a shared reference image. Anyone can use the same image, reconstruct it with any model or pipeline, evaluate the result with WOMetrics, and try to beat our score. **Beat WOMPar. Go under par. Advance the frontier.** **Navigate — PrismDeck.** Our reusable, intuitive, skeuomorphic HCI organizes images as parallel planes within a continuous 3D temporal prism, connecting memories, worlds, movies, and keyframes. We even use it to present our own pitch deck. **Built from Scratch in Four Days. Informed by 16 Years of Intuition.** WOM was created by an independent AI3D researcher with ACM SIGGRAPH publications, the SIGGRAPH 2020 Real-Time Live! Audience Choice Award (DrawmaticAR), and a CVPR 2025 Best Demo Honorable Mention (AI3D Render). Codex, Kiro, Muse, and VS Code accelerated implementation; years of XR, AI, and HCI research guided what to build. Human memories can blur or change each time we revisit them. **WOM imagines the inverse.** The photograph remains unchanged, but as spatial AI advances, we can return to that same image and reconstruct its world with ever greater, measurable fidelity—distinguishing what the image preserves from what AI must infer. **The photograph stays. The world comes into focus.** *WOM — Explore old memories. Create new ones.*