Raku Capture turns a phone scan into a Gaussian splat your AI assistant can talk to — then gives you a shareable link so anyone can view it, embed it, or drop it into their engine.
Free during beta • Phone or prosumer camera • No account required to try
The capture-to-asset foundation is live in beta. Steps 1–3 work today; the AI editing loop in step 3 deepens in Phase 2.
Walk a phone around a space or object. Raku Capture records the frames your phone or prosumer camera can provide — no scanning rig, no specialist software.
Our pipeline converts your frames into a Gaussian splat (.ply / .spz), normalized to engine-sane scale and coordinate frame. GPU-backed reconstruction; quality scales with hardware availability.
Publish to your creator profile and get a shareable link. Anyone can view the splat, embed it, or export it into Unity, Unreal, or WebXR. Your captures are yours — export is always available.
A published, READY capture resolves to a real, downloadable export for each engine target — the genuine reconstructed splat artifact already on your asset, packaged with an honest import hint. Nothing is re-generated on the way out.
Splat formats (.ply / .spz / .sog) export as delivered for every target below. Where an engine wants a mesh (.glb) or USD instead, that conversion is a gated capability — we tell you it is required and name the missing pipeline rather than handing back a fabricated file.
Loads the splat URL directly in the RakuAI WebXR viewer — it ingests .spz / .sog / .ply natively. No conversion, no download step.
Import the .ply / .spz artifact with a Unity Gaussian-splat package (e.g. UnityGaussianSplatting). A mesh artifact, when present, imports through the standard model importer.
Import the .ply artifact with an Unreal Gaussian-splat plugin. A mesh artifact, when present, imports through the standard FBX / glTF path.
The redesign step — capture an empty room, ask your AI assistant to stage or restyle it with Meshy.ai generated furniture, then export the dressed scene — is wired as a seam in the export path, not yet a live feature.
It stays honestly gated until the Meshy partner key and generation credit land. Until then the seam reports its gated state — it never returns a generated asset it did not actually produce. Your capture still exports today, exactly as scanned.
Try the path on a real capture — browse a published splat or scan your own. Export resolves the genuine artifact on a READY asset; before reconstruction finishes there is no export to hand back, and none is invented.
Every published splat is also an MCP-drivable scene. Connect Claude, GPT-4o, or any MCP-compatible model and issue natural-language commands against your captured space — 17 native tools plus passthrough access to ~9,500 API endpoints.
Deep scene editing (object replace, relight, place/remove) is a Phase 2 deliverable, not fully shipped today — but the MCP connection and basic tool calls work now.
Explore the MCP surfaceraku.scene.place_assetraku.scene.anchorEditing a captured scene with your AI assistant is graduating from demo to reliable. Here is the honest state of each refine operation — what works today versus what is still hardening in Phase 2. We label it straight so you can plan around it.
One rule holds throughout: nothing edited or generated is reported as exported until that loop is actually live. The export path stays exactly the contract above — native splat now, mesh/USD conversion gated. We never hand back a scene we did not really produce.
Connect Claude, GPT-4o, or any MCP model to a published scene and issue basic tool calls — query objects, inspect the splat, read scene state. The 17 native tools (5 read + 12 mutation) are live now.
Object replace, relight, place/remove, and undo exist as MCP tool calls but are still being made reliable. Treat deep edits as in-progress, not production-stable, until this graduates.
Generating photoreal furniture and restyling a room via Meshy.ai is the same seam from the export path — honestly gated on the partner key and generation credit. Until those land, the call reports its gated state; it never returns a generated asset.
Batch export presets per engine (Unity / Unreal / WebXR) and the end-to-end Meshy redesign loop are the Phase 2 finish line. They ship when they are real — and you will be able to tell, because the export still resolves only the genuine artifact on a READY asset.
Every published capture adds to a growing corpus of real-world spatial assets. More captures means better reconstruction priors; a richer library attracts developers who embed and link back; links bring more creators.
Phase 2 wires in Meshy.ai photoreal furniture and room-redesign generation. Capture an empty room; issue an MCP command to stage or redesign it; Meshy-generated assets are placed by the AI into the splat scene and exported engine-ready.
The partnership direction is real and scoped. A fully shipped end-to-end redesign loop is a Phase 2 deliverable — not a current feature.
Phase 3 adds a publish/consume marketplace: asset versioning, licensing metadata, and provenance on every published capture. Creator payouts are gated on real marketplace volume and GA pricing.
There is no payout system today. Your captures are always exportable and always yours regardless of marketplace status.
Capture is free during beta. Splat output in .ply and .spz; shareable profile link; export always on. Early-access members get first access to Phase 2 MCP editing and the Meshy redesign loop when they ship.
Beta • Phone / prosumer-grade reconstruction • High-volume GPU throughput gated on cloud credit approvals (Azure A10, AWS G5, Nebius H100 — expected weeks)
This creator hasn’t published yet — spatial captures drop here as they’re processed, so check back soon. Want to publish your own? Start capturing with Raku Capture.
Loading profile...
Want to publish your own spatial captures? Learn about Raku Creators →